Engineering types and power users need not apply,but For Everyone Else, the "normies", Qwen Flash Next is probably "640k is enough for anyone" territory.
IPO, valuation etc depend heavily on frontier, or "good enough" AI to need to run on commercial grade GPUs. If Qwen Flash Next is good enough and runs on a $2000 GPU, why would you pay $20/30/100/200 per month in perpetuity?
I wouldn't pick Qwen Flash Next to add kernel namespacing and cgroups v2 to my custom OS kernel, but it will happily implement the Plan.md that Fable 5.2 wrote for a dollar.
You can read the WSJ article
https://archive.is/20261113091227/https://www.wsj.com/tech/j...
And my comment in the other thread
It may not be illegal but if you are investing in AI companies you need to think about whether the demand for AI chips is real or artificial and being manipulated by Nvidia.
https://news.ycombinator.com/item?id=50493251#50496483
> Then he went on to describe the situation the company was in. Major cloud providers were running away from Nvidia, making plans to use alternative chips and lessen their reliance on the firm. Even worse, Huang said, the vast supply of chip wafers and memory Nvidia had committed to buying were coming at him like an “avalanche.”
Jensen's son-in-law was making these vendor financed deals and Kress the CFO now interim CEO was against them.
Large customers like Nvidia and Apple want billions of chips. TSMC can't just ramp up and down instantly. It takes 3+ years and $30 billion to build a modern fab. So TSMC has contracts where you are commiting to buy the wafers for years in advance.
I've worked for multiple fabless chip companies that use TSMC to manufacture. Amazon, Google, Microsoft, and the other big hyperscalers are making their own AI chips to move away from Nvidia. The startups like OpenAI are doing the same so the real demand for Nvidia chips is dropping. But Nvidia can't easily get out of the contracts with TSMC to manufacture and buy the chips. They would probably have multi billion penalties that would show up on their financial statements.
Some people think Nvidia is acting as a VC fund for AI startups but my view is that Jensen freaked out about the dropping demand for Nvidia chips. So he is funding these startups with circular finance deals to buy the Nvidia chips that he can't sell to the hyperscalers anymore.
But weren't they were investing money in companies to help them grow and in turn buy more of their product. Of course Nvidia wants to stimulate demand of their product. I don't think this was nefarious. There is no free lunch, the risk of them never getting that money back is real with a lot of these pie in the sky AI companies. How is this anticompetitive? It's all of these SPVs that meta, google, and others are using that I think needs to be investigated and punished by wall street.
If you think that's wrong, USE YOUR WORDS, I'm open to reason. Down voting with no response or making jeers is not effective communication.
Nvidia allegedly engineered scarcity by funding AI startups to buy its GPUs (techpowerup.com)
Greed and hubris. I think the turbulences will be painful to some, especially those who counted on Nvidia stock to finance their retirement.
But, big picture, this is good news.
L3Harris has been modernizing the communications infrastructure supporting the National Airspace System in close coordination with the FAA. With safety and reliability as our highest priorities, L3Harris, together with its telco partners and the FAA, has completed more than 70% of the planned airspace communications conversion from legacy copper wires to a fiber optic system. We are supporting the FAA in its investigation to understand the conditions that led to the Nov. 13 ground stop of our nation’s air travel. L3Harris takes the nationwide ground stop and its impact on travelers, airlines and air traffic operations seriously.
<end entire statement>I had thought the L3 was from Level 3 Communications and Harris, but its actually L3 Technologies which seems to be a holding company for defense contractors.
Great first impression from a technical company
Thank goodness
> Agent messages
> Your AI agents can message you inside the game. A message costs the agent only the words in it, and nothing prints on success. Messages wait while the game is closed.
It's kinda amazing how the ancient "local mail" technology keeps getting reinvented, you know, and reintegrated inside the gameplay? NetHack had its its "mail demon" since... (googles) version 1.0.2, in April 1985 — the creature would spawn in the game, deliver you a scroll of mail, and vanish. And apparently, it's still in use nowadays? "On public servers such as nethack.alt.org, a user may observe other users' games and communicate with them by sending mail."
I took over an abandoned daytrading sub on Reddit. I think when I took over it had something like 30k weekly visitors.
It basically had no rules, anyone could post whatever they wanted. It was mostly ads, funnels to some apps or YouTube videos, courses, private chat groups, and so on.
I decided to do a cleanup and ban all the cross posts, obvious AI posts, bots, and ads. And it killed the entire sub, it had zero legitimate posts after.
Makes me wonder how much of LinkedIn is actually just fake jobs, either MLM or ghost jobs, etc.
There are levels to the game everywhere. One example that I can think of is on the timeshare stuff in Mexico where they exploit retired older people into buying fake ownership of something and end up instead just owning a huge debt.
It’s like the world has entered an era of abundance where the game is no longer about value but rather perception of value and ways to pressure your psychological makings.
I show up at a building 2 blocks from school, a random tenant tells me to form a line by the door that was locked. I was confused because I was the only one there. A second person came in, joined the line. Then a third, fourth, fifth. We were a crowd, standing in front of a door, where I thought I'd interview for a entry level job in tech.
When the door opened and the person on the other side greeted us and a good number of people immediately left. More luck for me to get this job. But turns out, it was going to be a group interview. In fact, not exactly an interview. It was a pitch. And a few lucky ones would have the privilege to sell high tech knives. All we had to do was buy the first batch of knives for something like $300.
Selling people mountains of cheap garbage that will break almost immediately is not much different than theft to me. But it's a scam that benefits enough other people like the shipping industry than nobody will rock the boat.
The economy in general is trending more into figuring out how to siphon wealth from people without money into the pockets of people who already have everything. Poor people are actually pretty profitable because they have little choice in spending.
And that's not even considering perfectly legal "scams" like shitcoin pumps and politicians awarding shady contracts to buisness people who donated to them without a technical bribe happening. I would love to know how much of the actual economy is above-board business and how much isn't, but the older I get the larger the "isn't" seems to grow. Is the scam-economy 1% of all real financial transations? 5%? 10%? 20%? I'm honestly not sure.
Anyone who thinks we’ll have data centres in space before we have a civilisation changing step up in surface to orbit space drives and materials science is an idiot.
How the hell does anyone think a rocket is ever going to compete with a prime mover towing two trailers, or an ocean going vessel.
We’re not exactly short of land.
And no, space is not cold.
If you want a safer datacenter you put it underground and you can safely assume that if it's hit by a bunker buster you have much much bigger problems.
For those reasons alone, and because compute will become more and more strategic for defense, I find it very hard to believe that data centers in space will represent even 5% of all compute in the next 10 years. I think a higher proportion will rather be underground or in bunkers than in space.
If that workload does expand in usage, and the economics of orbital DCs make sense (with high launch costs, zero ability to do maintenance, and weight penalties for shielding and radiators), then surely ocean based data centers (e.g. Panthalassa) make more sense (lower 'launch' costs, high but existant latency on maintenance, no weight penalties), and if those make sense then surely the distributed land based facilities (e.g. Rune Energy) or home based facilities (e.g. Span XFRA) make even more sense (easier maintenance, low/no permitting overhead because they're covered by existing permits, reasonable quality terrestrial fiber internet).
If you, dear xhevahir, wanted to buy a math phd attempt and had the chops, you could self-fund the training, courses, and examinations of a Math PhD, and it'd be up to your own abilities to finish.
But nobody, nowhere, would (should!) stop you because "the jobs" aren't there. You could do it because you can find the money to do it.
Asking universities to set their intake numbers according to the perceived demand 5+ years down the road (let alone the coursework + specialization of those PhDs)? Madness - they'd effectively be planning the whole economy at that point.
> “Stipends, insurance and the five-year funding guarantee all remain in place for students making satisfactory academic progress,” Colla wrote. “Advising relationships, qualifying exams, degree requirements and teaching and research appointments will continue as planned.”
What a huge goddamn waste of time that was.
It got to a point that to preserve my sanity at work, I kept my head down, did the work that needed to be done (whether assigned or not), stopped going to meetings I didn't care about, and stayed mostly silent in all others. All the while dreaming of a new career or early retirement. But then a strange thing happened... It turns out that if your reputation is one of quiet confidence, people start to actually listen. They assume that when you do talk, it's only going to be about something important and worth hearing. Instead of co-workers on other teams trying to throw their work over the wall at you, they begin every innocent question with an apology: "hey, I know you're super busy, but I was wondering if I could bug you real quick..." Borderline indifference is so far my favorite superpower that I've accidentally discovered.
Long-term, it can be hard to sustain, and maybe a bit unhealthy – it becomes a form of self-betrayal, where you repeatedly suppress genuine negative feelings about dealing with someone difficult. These feelings are there for a reason, even if expressing them with the person triggering them would be counterproductive.
Also, in more extreme cases, the sort of people who are prone to creating drama and conflict pick up on the technique and make it into a battle of wills – who will break and give up first. And they may be a lot better at it than you are, because for them it is a central feature of their personality, and for you it's just a tactic for immediate relief.
Grey Rock method worked against a particularly toxic work conflict that was keeping me awake at night... but... the reality was that it produced zero work results for any of us.
Spoiled Americans. That drive is $1393 (14160 SEK) here in Sweden now[1]. A year ago it was 4160 SEK...
Three times as expensive in 12 months seems true for most drives.
I'm just praying no drives in my NAS are going to crap out anytime soon...
- ARC may require manual tuning for size and metadata size to accommodate page cache, in-app caches, file count etc
- L2ARC and SLOGs are very obviously useful for some clear workloads, but have some cliffs you don't expect, such as ram required for L2ARC mapping, unexpected cold/hot/boot behavior
- Reasoning about reliability becomes harder the more tiers you have
Just imagine, to pick a simple example, you're running postgres, thousands of sqlite3 dbs and one huge duckdb (larger-than-ram) on a single such system - it would perhaps manage to adapt to one such workload at a time, but would it balance all three at the same time?
So much as I'd love to see this direction, I can also imagine the pains coming with it.
Building: the videos/tutorials from Andrej Karpathy in Python [3] are considered to be very good at building and understanding the basics, from a small/toy neural network engine to transformers and tokenizers.
There was a post here a few weeks ago [4] on reasoning models using Python/PyTorch with links to an older series on non-reasoning transformers.
[1] https://www.youtube.com/playlist?list=PLZHQObOWTQDNU6R1_6700...
[2] https://www.youtube.com/@WelchLabs/videos
It’s a 36.2M-parameter version trained on a MacBook Air M5 (24 GB) with jax-metal, on 207 old detective novels. 12,000 steps took about 2.6 hours.
Full write-up with diagrams and results: https://vaibhaav.in/writing/deepseek-v3-jax/
The trained checkpoint and exact data are in the v1.0 release, so you can generate text without training.
Would love corrections if I got anything wrong :)
Did you check with internal counsel?
> There has never been an illusion that anything said, written, or done at work is private from the company
The entire industry is one court case away from banning transcription. Not many companies want every word of every meeting legally discoverable - in the past, meetings were used to avoid having a written record. The implications of ubiquitous transcriptions haven't percolated yet, I give 1-3 years (or until the next major case)
Before I turned it on I asked my team if they had any objections. Silence for eight seconds before saying if anyone was uncomfortable objecting in the group they could reach out to me individually. No one ever did. The only request was that everyone internal get copied on the notes so they could correct them if needed.
These are company decisions to have these connectors and privacy choices. It's their data and their risk. There has never been an illusion that anything said, written, or done at work is private from the company. At home I partake in none of this.
That’s a world I don’t want to endorse, but it’s seemingly impossible to opt out and stay employed. Is there anything that I can do to minimize my exposure? Has anyone made it a norm to say “transcription isn’t needed for this meeting so I would like everyone to turn off AI notetaker before we proceed?”
This is a quote from an internal Meta memo about how they would plan the launch of these creepy glasses. There are too many crises that seem more pressing to a lot of people.
"Article 12
No one shall be subjected to arbitrary interference with his privacy, family, home or correspondence, nor to attacks upon his honour and reputation. Everyone has the right to the protection of the law against such interference or attacks." https://www.un.org/en/about-us/universal-declaration-of-huma...
"arbitrary interference with his privacy"
That is tech companies does not have right to private data through data fusion. We humans collectivly have the right to privacy.
It is a bit more complex because we have federal and state taxes but there really is no reason why we couldn't also have that system.
Twitter, as they say, is not real life. It's just a broken clock that might be right twice a day by happenstance.
Like, that meme was astroturfed right? People in Europe don't talk like that or act like that.
And we also had foreigners from all around the world (Japanese, Korean, etc. etc.) come and be all Costco is amazing or other tidbits of pro-Americanism.
I dunno, something just felt off by it. There's clearly something going on and it doesn't feel completely natural.
It's fine to like America and be patriotic and all. But for it all to strike at the same time with extremely similar messaging? Even memes had more diverse concepts than what happened.
We are IMO in the age of AI boosted memes where people are trying to form large scale reality distortion fields.
> I think it's clear consent was given at some point
That's not what was found. I don't know the facts of the case, but it's possible that "consent" mechanisms were found to be insufficient. As an example, if you ask for permission for one purpose, then use it for another, that would probably not meet the "inform users" bar.
How does this younger cohort feel about locked down devices, App Store distribution and the like? In my experience younger people don’t think about it at all, because they aren’t aware of what the alternative could even be.
I’m in my 40s and have been working in tech for a long time. I remember a time when it genuinely felt like the industry was a force for good in the world, opening doors for people to do things they never used to be able to do. And the companies were so open, e.g. I remember Facebook’s OpenGraph opening up a world of new social apps that were exciting to develop and to use. For sure there was a level of youthful naïveté at work there but it was still an exciting and interesting time to be in the industry.
That positivity still happens today but to a far less degree. Nowadays it feels like every major tech player is all about lockin, about dark patterns that draw users into their services whether they’re happy about it or not.
We used to point and laugh at the cynical, out of touch companies like Oracle. Now it seems everyone aspires to be them. So yeah, I’m often negative on HN.
Or leave the market altogether. Society here would be better off without Meta, for sure.
TL;DR: they didn't thoroughly explain the users what they were consenting to when giving the app access to the gallery.
Google hires some hardware people to come up with general requirements for specialized stuff like ASICs, then sends those requirements to Broadcom to do the hard parts and negotiates big multi-billion dollar contracts with them. OpenAI did the same and contracted Broadcom after liking what they saw with the TPUs. There are several other Broadcom customers that I'm probably not at liberty to disclose.
But no, as far as GPU replacements are concerned, the Hyperscalers generally are not experienced enough in the specific types of hardware required to do this by themselves. It's very specialized and boutique high price work and more difficult than say, designing an ARM chip (something a company like Amazon or Google could probably handle in house).
In general though, yes, I agree. Hyperscalers are trying to reduce dependence on NVIDIA. Many are quite far along in their ASIC projects (Google especially, they are on their 8th generation TPU! Way past just being a small scale experiment by now.) and have large volumes of them running in place of NVIDIA chips and much more efficiently already.
Broadcom is fabless, and uses TSMC for these sorts of chips. They are mostly a boutique design "mercenary" firm for exactly this sort of specialized hardware for Hyperscaler customers and other clients that run significant amounts of infrastructure in house (so, think JP Morgan for instance). That's the core of their business, along with increasingly Infrastructure Software in the form of virtualization and mainframes and private and hybrid cloud (so some natural sales overlap there, just generally, "big companies that need specialized big data center hardware and software").
The WSJ article is basically saying the opposite. The hyperscalers are making their own chips and they are buying less from Nvidia.
https://archive.is/20261113091227/https://www.wsj.com/tech/j...
This is my comment in another part of this thread
https://news.ycombinator.com/item?id=50493251#50496483
> Then he went on to describe the situation the company was in. Major cloud providers were running away from Nvidia, making plans to use alternative chips and lessen their reliance on the firm. Even worse, Huang said, the vast supply of chip wafers and memory Nvidia had committed to buying were coming at him like an “avalanche.”
Jensen's son-in-law was making these vendor financed deals and Kress the CFO now interim CEO was against them.
Large customers like Nvidia and Apple want billions of chips. TSMC can't just ramp up and down instantly. It takes 3+ years and $30 billion to build a modern fab. So TSMC has contracts where you are commiting to buy the wafers for years in advance.
I've worked for multiple fabless chip companies that use TSMC to manufacture. Amazon, Google, Microsoft, and the other big hyperscalers are making their own AI chips to move away from Nvidia. The startups like OpenAI are doing the same so the real demand for Nvidia chips is dropping. But Nvidia can't easily get out of the contracts with TSMC to manufacture and buy the chips. They would probably have multi billion penalties that would show up on their financial statements.
Some people think Nvidia is acting as a VC fund for AI startups but my view is that Jensen freaked out about the dropping demand for Nvidia chips. So he is funding these startups with circular finance deals to buy the Nvidia chips that he can't sell to the hyperscalers anymore.
Sure they have customers pounding at their door for more product. At some point the hype machine stops and new neocloud startups can't get funding because its too expensive or banks don't trust them. Then what happens? The circular financing starts to unwind itself and NVidia is the "bag holder". If you look up Lucent or Nortel this happened to them during the dot com boom. There is a positive difference here with NVidia, they have so much demand and usage this can probably go on for a while. Lucent and Nortel were bit by dark fiber and the usage/demand just wasn't there.
If Jensen at the helm of NVidia is bullying/convincing companies into purchasing more than they need, financing it, etc. and creating the hype machine himself and NewCEO ain't going to play that game, then it could be an end to the circular financing.
‘Circular financing’ sounds sinister but customer financing and vertical investments aren’t strange or uncommon. It could be bad for sure. But it is not like they invented a new form of wickedness.
“He’s a better businessperson than all of you,” Huang told the cloud computing CEOs. The soup-man doesn’t sell out all his inventory to one or two customers, but instead keeps prices high and his customer base diverse, the CEO said.
“You are not establishing the value of my products,” he said.
> Then he went on to describe the situation the company was in. Major cloud providers were running away from Nvidia, making plans to use alternative chips and lessen their reliance on the firm. Even worse, Huang said, the vast supply of chip wafers and memory Nvidia had committed to buying were coming at him like an “avalanche.”
Jensen's son-in-law was making these vendor financed deals and Kress the CFO now interim CEO was against them.
Large customers like Nvidia and Apple want billions of chips. TSMC can't just ramp up and down instantly. It takes 3+ years and $30 billion to build a modern fab. So TSMC has contracts where you are commiting to buy the wafers for years in advance.
I've worked for multiple fabless chip companies that use TSMC to manufacture. Amazon, Google, Microsoft, and the other big hyperscalers are making their own AI chips to move away from Nvidia. The startups like OpenAI are doing the same so the real demand for Nvidia chips is dropping. But Nvidia can't easily get out of the contracts with TSMC to manufacture and buy the chips. They would probably have multi billion penalties that would show up on their financial statements.
Some people think Nvidia is acting as a VC fund for AI startups but my view is that Jensen freaked out about the dropping demand for Nvidia chips. So he is funding these startups with circular finance deals to buy the Nvidia chips that he can't sell to the hyperscalers anymore.
I don't deal with their big customers, but the idea that NVIDIA doesn't have customers ready to spend $$$$ to get some GPU in their datacenter is proveably false. It's likely easier for them to handle big transactions for big customers, but there's also a loooong tail of customers in line for a smaller piece of the pie.
This scheme had pissed off bankers that a company muscled in on their turf, created lots of debt in a time when US debt was already sky high and becoming unpopular, relied on GPU depreciation being low (implying tech progress was slow, which didn't gel with the singularity crowd), and getting crazy rich in the process.
I don't even know how China factors into this other than competing with the US
Because of bottlenecks in the supply chain, Nvidia knows it is going to sell 100% of its inventory. They could raise prices quite a bit more, but then their only customers would be hyperscalers (building competing chips) and possibly OpenAI and Anthropic. Nvidia is willing to forgo present day profits for the strategic value of keeping its customer base diverse and promoting the network effects of e.g. CUDA. Instead of quoting a lower price per GPU to neoclouds and startups, they finance purchases of chips, but the two are roughly equivalent (financing means Nvidia takes on more risk, but has more upside and locks in future demand). It is in their commercial interest to have many competing buyers of their chips.
Obviously this creates fun mechanisms for fraud and wacky accounting (and I'm agnostic as to whether that's happening here), but it's worth knowing about the boring/obvious story before donning your tinfoil hat.
(circular financing)
>Huang had launched a dealmaking spree and was now starting to question whether it had gone too far, according to people familiar with the matter.
(neopotism, Huang’s son-in-law, Nico Caprez)
>Huang’s son-in-law Caprez and another vice president Raj Mirpuri told prospective business partners that Nvidia could arrange credit-support deals within a couple of weeks. Dozens of companies lined up for this program, hoping Nvidia’s backing would lower their borrowing costs. In exchange for providing credit support, Nvidia proposed that cloud providers share a cut of their revenue.
>Behind the scenes, she [Kress] tried to rein in Huang’s son-in-law, who had been promoted to a vice president in January, people who worked with her said.
(...)
>By October, Nvidia started to turn off the tap.
> [Story about Japanese Soup vendor] The soup-man doesn’t sell out all his inventory to one or two customers, but instead keeps prices high and his customer base diverse, the CEO said.
ORGINAL is the best. It tells completely different story better (not a weird twist where CEO trying to increase sales and price of their product is suspect)
Before His Ouster, Jensen Huang Belittled the Executive Replacing Him --- Interim CEO Colette Kress has questioned some Huang dealmaking. He called her ‘a deer in the headlights.’https://wallhop.io/https://www.wsj.com/tech/jensen-huang-col...?
https://wallhop.io/https://www.wsj.com/tech/jensen-huang-col...?
As you say, if Nvidia is chastened enough to slow its banking endeavours and that lets air out of the bubble in a progressive rather than dramatic way, that’s probably better for everyone (except OpenAI, who are probably hosed in that scenario, as they have huge commitments but are struggling to raise what they need; Anthropic might last long enough to reassess)
Connected. Complete lack of integrity.
If he cheats on his wife, and on his customers, he'll probably cheat on his shareholders. Hence having to quit before he's pushed.
Huh. The circular financing was figured out a while back, but this is something I had not seen before.
This could all work out. NVidia stock drops. AI company stock drops. The more efficient approaches to inference and more open models take over for most mid-level applications. The need for insane numbers of data centers is reduced. AI works fine and costs less.
I also had the displeasure to work with nvidia. After years of pushing gaming cards on us for compute, the years I spent working with nvidia was full of conversations where they would list silly excuse after excuse as to why we should use the expensive enterprise-grade cards instead. All the while, we begged them to unlock driver features that were available to Quadro cards but not GeForce cards, features as simple as hardware acceleration pass-through on a remote desktop connection. It took them a full-on epidemic, where remote work became super common, to allow some of those features to GeForce cards.
So it's really on brand that they tried creating a fake scarcity to keep their prices up. I don't envy the situation Jensen is in but I won't be sorry for him.
Nvidia gives Nebius first access to their newest chips, ensures they get deployed at top performance, allows Nvidia to recognize the revenue upfront, and ends up owning equity in the upside.
Nebius gets the best chips first, gets cheaper financing due to the obvious security granted by having one of the most valuable companies in the world standing behind them, etc.
It sounds like a mutually beneficial deal rather than something shady to me.
By the way they report demand is 10x supply so I think everyone in here hoping for some cheap GPUs soon is kidding themselves. Eventually RAM might get cheaper because it's easier to build new factories to produce commoditized RAM but TSMC is the bottleneck for advanced chips and it isn't resolving anytime in the foreseeable future.
I'm buying the dip from this silliness.
Since these firms (neoclouds and AI startups) will get consolidated if Nvidia puts too much money into them, I assume the majority of their funds are coming from elsewhere.
https://www.theguardian.com/technology/2025/dec/28/nvidia-in...
The internals are quite interesting, I always assumed the interpreter just processed the input as it arrives but it actually starts a PostScript program as soon as the printer boots and that controls the hardware and handling incoming jobs. A lot of this code is available through a console, ie:
userdict begin /start load == end
dumps the first program started by the printer.I have an original LaserWriter which works through the console but doesn't print. Luckily someone has done the work to get the firmware running in MAME.
My interest in this came from working with a QMS distributor in the eighties. At the time I didn't really have a chance to delve that deep into the technology but I was always fascinated by it.
Mainly I learned that Ghostscript is doing a lot to make its PostScript interpreter fast: with the non-graphical test programs I could think up, GhostScript was many times faster interpreting the program (even computationally intensive ones) than my runtime (called "egon") could execute the byte-compiled PostScript. So it was a computation fail but still a pedagogy win.
I've always been meaning to learn more about how Ghostscript is engineered.
Did I get that right? Seems a bit anti-pythonic.
Like, if I want to host on Codeberg, or Sourcehut, what is the process needed for those platforms to become a trusted publisher?
As it happens, I build my releases on my own hardware using a pypa/manylinux release. Can I provide attestation, or am I required to use a third-party service to get that "x Without provenance" removed from my entry?
A comment from age 77: the single thing I am most grateful to have retained at the same intensity it was in my youth is my curiosity.
Most people, even technical people, can't imagine being 90 years old and being indistinguishable visually and in terms of mental capacity from a 20-year-old. They imagine decades of suffering with bone aches, diseases and dementia... even technical people who should know better that when you optimize a system that's full of technical debt, it's almost all upside.
I'll have plenty to do and plenty to think about when I'm 100 years old, assuming I have a body and brain capable of it.
"Yes, but what about all the people you'll have lost by then?"
First of all, if I get to live a long and healthy life they probably do too.
Second of all, we adapt. It's not all sunshine and rainbows, but getting to live for as long as you want, as healthy as you want, is worth the grief.
Granzyme K is the protein.
Jon Blow doesn't have sufficient experience in using LLMs to talk about it.
This is evidenced by his focus on individual anecdotes on AIs failure to deliver something.
I would love to talk to him about this as I have fairly concrete evidence of the insane productivity gains one can get with smart AI usage.
Also would love to hear Ryan J Fleury talk about Tim Sweeneys statements on LLMs.
How long until RADDBG is primarily used by Claude? How long until Ryan is told he must support this? How long until he claims he was never against it?
I regularly meet (recently examined) people who are trained "software architects". Common among them is that they are often delay projects championing their pet idea, sometimes bringing negative value to the project. The best programmers are just so much better software architects also. So I expect history to repeat at this point.
The systems analyst analyzed the problem that needed to be solved and designed a solution involving a computer, including system modules, inputs, outputs, etc.
The programmer, supposedly a lower paid profession, transformed that into Computer code. I remember someone telling me that wasn't the best choice and that I should become a "systems analyst". They wore the better suits.
My college even offered a shorter, cheaper degree for people who just wanted to be "programmers".
I think this distinction still survives in government labor statistics.
Perhaps we came full circle with LLMs partially killing the "programmer" job.
We need more content creators focusing only on their field of expertise. I don't care what Jon and Casey think about React or big tech or coding interviews. But I find their gamedev experience and insights valuable.
And for the pixel perfect complaints, it only shows lack of care to not have learnt how layout managers work.
People there will cheat every way possible to get their framerate/performance/cpy cycles. Yes, it's often amusing to read about really clever and interesting approaches, hacks or even framings of the problem. Like even one of Blow's presentations, where he was explaning how Braid did time reversal. Naive intuition tells: "make everything deterministic", Blow's answer: "how about just tracking state of everything".
But gamedev is not "orders of magnitude complex than any other software". State of the art databases need to be fast and correct. Even that alone makes them far more complex than "state of the art games".
And I'm sure there are many other areas who will look at databases and also rightfully tell that they are even more complex.
I sometimes wonder if I'm a victim of gel-mann amnesia when I listen to his opinions on gamedev too :(
The changes always come in monster one-shot PRs ofc - too much for a human to read - so I just feed the branch to my Claude (identical to the one they’re working with) and after some thought it will give me 15 to 30 suggestions of what needs fixing.
I’m still not understanding why this happens?
The latter is why managers love AI - and the former is why the general public hates it.
We're accelerating our ability to produce mediocre software that solves common-place, mediocre problems, but with more "freedom", more customisation.
His Lordship Mr Blow, has something of an academic sense of what innovation means. He wants to see new problems being solved. A new O/S that is "secure by design" in throwing away old models of shared-CPU schedulers, and the like. I assume, new techniques in 3D game graphics. New ways of mixing compile-time/run-time langauges to accelerate development. And so on. New in the way that an academic research project should be new.
What we have from AI in software isnt anything new. Claude releases an office suite with AI assistance. We havent yet seen AI accelerate anything other than end-user programming, or lift-and-shift projects.
These might be very valuable, but theyre not innovative.
A year to ship is the old paradigm. With 100% agentic vibecoding, everything should be amazing now.
What will next year bring?
For WiiU, Nintendo used the Green Hills compiler. Their entire toolchain was simply clunky and unstable. You had to use this MULTI debugger which often didn’t work. My favorite line from their dev docs was “c++ not fully supported”. Well, that’s an issue since we used that extensively. It wasn’t even clear what actually wasn’t supported.
He regularly denies the use of new technology that is genuinely useful, and is a fairly conservative dude (I mean in the non-political sense. I do not know his politics. He may be a conservative politically as well, but that would be independent of the point I'm making.)
I've worked with many engineers at many different scopes. Big companies, indie darlings, web services, games. Games is always the last to adapt, always the last to think that anything they didn't invent is worth their time. Blow has experience in that domain, and is colored by working in it, but I don't think he knows what's happening on the bleeding edge very well.
Their workflow was:
1) Arrive at work. Get coffee. Fetch latest code from Visual Source Safe.
2) Turn on Xbox. Hit F5 to build, deploy and start debugging on target.
3) Edit and Continue C++ in debugger IDE all day on target without ever bothering to stop the game. (Yes, this took some coding discipline to maintain.)
4) Turn off Xbox and go home.
Between the Visual Studio debugger and the Whole Tomato coding assistance, it was easy to understand the code and the runtime behavior. For anything GPU-related, Microsoft Pix for Xbox gave you a scrubbable timeline of every action, every pixel, every register change in the GPU while drawing a frame.
The hardware capabilities, available libraries, and artist tools have increased tremendously since then. But, the coding experience has not. 20 years later, I'm outside of gaming and I find most people using re-run-and-printf debugging for professional development. Even locally. I don't know how anyone gets anything done.
The only significant improvements I've seen in 20 years (besides the occasional memory tracker or profiler tool) has been AI agents and reversible debuggers. Otherwise, it's all been downhill since 2004.
1. Hardware advancements mean that you can get the same results with less optimized code (this is what allowed 3rd party engines with GC scripting languages to become popular)
2. The proliferation of learning resources and open source code.
3. The budgets and size of indie studios as well as total time spent on the games. Seriously look at some of the games these days that get labeled indie.
Warcraft 1 (my first RTS) had just 4 programmers. And only one of them was on the project the whole time. And the entire project was only 14 months.
Manager. FAANG. Enjoys the misfortune of the people he manages. Your friend sounds like a prick to me.
In the last 3 years, there have been a lot more indie games of far higher quality than we used to get. I'm thinking of the RTS space right now. The Scouring for example. We went from having nobody but Blizzard able to make something that complex to several games with outstanding performance and complexity, often with their own custom game engines that are not only working but working better than I remember wc3/sc2 working at launch.
I don't know how much of a role AI is playing in those games, but I think the performance and complexity of indie games is shooting up and we are seeing very, very small teams make things that large teams struggled with in the past. Honestly, I think indie studios are outperforming AAA in most ways besides spend and length and art budget --and yet, it's the indies who are making their own game engines these days, isn't it?
(btw, we have to also admit that many of these game engine builders look to blow for inspiration (to carmack too), and their architectures are not what you'll find at typical saas companies. Game coding is complicated in a really fun way I didn't get doing plumbing on web projects, speaking for myself)
And I agree, it’s because organizations and people haven’t changed (yet, if you’re being positive) what it is they’re optimizing for and how they do work. AI gives you more of what you want and gives it faster, that doesn’t on its own change the culture or goals of a company.
It would be interesting to require laws to be compilable. Charles Stross wrote a couple of books that include that idea.
At a high level, the core is 2-wide fetch/decode/rename/retire with a 128-entry ROB, 160-entry integer PRF, 8 branch checkpoints for single-cycle speculative RAT rollback, and separate Int/Mem/FP/Vec issue queues. On the memory side, it has a dual-port Sv39 DTLB (64-entry 4-way Tree-PLRU + 8-entry 2 MiB/1 GiB superpage CAM), ITLB, and a 16 KiB 8-bank non-blocking L1D. Instead of flushing dirty lines on every page-table walk, the hardware PTW snoops dirty L1D sets non-destructively so Linux page-table updates are visible directly, paired with store-queue drain on SFENCE.VMA.
In Verilator, it boots mainline Linux 6.6.50 and BusyBox userspace in ~89.8M cycles (~76.2M retired instructions, 2,255 page faults handled) while checking an 8-bucket single-owner CPI conservation invariant every cycle with 0 cycle leakage. Synthesizing the 8 main core logic blocks standalone with sv2v + Yosys + OpenSTA on Nangate 45nm (SRAMs/PRFs black-boxed, pre-P&R) meets 500 MHz across all 8/8 blocks at ~219 kGE, after rebuilding the ALU and branch unit as Kogge-Stone prefix trees and balancing high-fanout CSR/TLB paths.
I used an AI-agent-assisted workflow for RTL iteration and verification scaffolding while driving the microarchitecture, timing closure, and debugging myself. The nastiest late bug was a Linux boot livelock around ~90M cycles: per RISC-V Priv Spec §3.1.9, CSR read-modify-write instructions on `mip` have to operate on the software-writable `mip_q[9]` bit rather than the live `irq_seip_i` wire—our CSR block was feeding `mip_val` back into RMW, so whenever SBI cleared `STIP` while a PLIC IRQ was high, it permanently latched `SEIP` and trapped the kernel in `plic_handle_irq`.
For folks who have built Sv39 cores or open-source ASIC flows: do you prefer letting the PTW snoop dirty L1D lines directly like this, or forcing a clean/drain path? Also curious what people would tackle next on a single-core tree like this—an inclusive L2 cache or jumping straight to dual-core coherency.
I can't even figure out if this content is supposed to be video, article or a microblog post. It just says "click the cover to read". I click things. Nothing happens. I don't know what the cover is or if I'm clicking it.
If you use the raspberry pi for homeservers or general compute (which I did/do sometimes) then the value proposition is quite poor regardless of pre AI prices unless you have very strict power requirements (in which case it probably fits the above use case already). The comparison with NUCs or mini PCs or laptop boards feels like it's a wrong comparison.
The chips used in PIs are repurposed Broadcom SOCs that have inferior bus speeds compared to a general purpose processor from intel or amd, even few generations old. For any general purpose computing these chips will always be slower. its like comparing apples and oranges. I never really understood why people did that.
The boards are of course bigger, which I guess will be a non-starter for many. They'll also require active cooling unless you downclock them considerably. They often also require more power, although many mainboards can be configured down two a few watts of power consumption, depending on what peripherals and speed you need.
Yeah, it's more work than just getting a Pi, but given the current prices, it makes sense for me.
I am by no means an expert at this, but starting of with a device on the Postmarket OS list and Claude, you can get a lot of these old devices working with mainline Linux and nearly all drivers. The newer devices are a bit harder, but if you go back nearly a decade, they're very easy to get working.
So far I have repurposed a Fairphone 3, Nexus 5, Galaxy S20, and Galaxy Tab 3.0 7". All of them (except the s20) with all drivers working compared to stock. Others have done a lot too: https://github.com/mlainez/awesome-nerves-systems
For DIY projects you can use host USB, or get one of the Adafruit FT232H type boards for GPIO pins.
Not my experience at all. Pies are hard to find, get more expensive each week. Some are impossible to find (Pi zero).
The most fragile part of a phone is the screen. If you don't care about breaking it, then access to the battery becomes pretty easy, and therefore easy to replace.
Used phones are good for tinkering, especially for "display with sensors and battery attached" usecases, but there are still plenty of downsides. And unfortunately Linux on phones is just not ready for primetime, last I checked.
Because the pi has a maintained kernel and a userspace I don't have to fight.
> The open question is supply. Finding two good used phones is easy; finding a few hundred matching ones, with healthy batteries and predictable camera behaviour, would take more work. I feel increasingly confident that it’s possible.
RaspberryPies and such are commodity hardware. You can buy one, two, ten, twenty, a hundred, and it just scales along. All of those devices will, barring perhaps a few dead-on-arrival ones covered under warranty, just work out of the box.
A used smartphone will have issues which can vary from device to device, so scaling that would take a very dedicated refurbishing setup where the quirks are ironed out (and the unsalvageable ones discarded).
I would also expect that you are starting out at the other end of the bathtub curve: with new devices you will have a few that fail soon after unboxing and using them a bit, and then they will run fine for years until you start hitting the other end of the curve and wear-related faults pop up. With old cheap phones you would start out at that side of the curve, so if you buy one hundred of those, a bunch will fail at various points soon after deploying them.
The cost of that unpredictable behaviour seems high to me. So much so that I wonder if it wouldn't be cheaper to buy a lot of unsold stock of a certain type of budget smartphone from two or three years back. Those devices could run fine as long as they aren't connected to the internet directly.
https://www.youtube.com/watch?v=aMct99DiZak
For my money, a triple burrito wrap of foil and ziploc gives 90+ dB attenuation plus moisture and mechanical protection.
Also nitpick, but the E1 pulse doesn't fry beefy transformers, instead it's the long-period E3 pulse (seconds to minutes) that raises the grid voltage and overvolts the transformers.
I don’t think the author realizes that “AP” is an exam for American high schoolers. Gates is only claiming he made new test questions at a high school+ level that wouldn’t have been in the training set for AI. I don’t see how it’s so unbelievable he could do that.
What's your definition of exceptional? Because that fits mine.
Being CEO does not involve a lot of coding. Yes he was an expert 40 years ago, but a lot has happened in 40 years
Around when AI was getting much better at SQL I was writing SQL everyday at work and had been for years. I don’t think it would’ve been hubris for me to judge that it was getting really good and would shortly be super human at SQL. It would’ve been correct anyway.
Then immediately goes on to say that his comments about AI coding are arrogant despite being an expert in that field with more experience than nearly anyone.
Unsure how to interpret that.
Eg I can put 2x of these workstations in an office handling banking/medical/secret/sensitive data and have my engineers stop complaining.
Edit: This guy already did it, and you get 60% more throughput on the 8xB6000 https://github.com/kacper-daftcode/GLM5.3-8xPRO6K
Why would any of those non-power-users buy a $2000 GPU that doubles as a space heater (seriously, try sitting next to one during long runs in the summer) and thousands more for a computer to host it, plus all of the trouble of setting it up, instead of a $20/month subscription?
I do have the hardware to run Qwen3.8-Flash-Next at home, and I still heavily use my subscriptions even for simple tasks. The experience of using the apps on my phone on the go is great and I always have the option to scale the model up to the task.
The people who are going to pay thousands of dollars up front to avoid $20 are power users and people who have different priorities and motivations, like independence or ultimate privacy. GPU prices dropping won’t change that much.
Also, you don’t need an RTX6000 for Q3.8FN. There are much cheaper options right now at lower speeds.