Where the AI market is talking

Anthropic Sees over $30T in Potential Revenue

The combined revenue of the SP500 ~$18.28T, or ~14.8% of global GDP in 2025 (~118.35T). So is anthropic really saying that they expect to have 1.64x the revenue of the SP500 or that the total addressable revenue for AI they suspect is $30T? That number is at least plausible but expect to earn 25% of global GDP is a bit whacky unless you've drunk a lot of koolaid

Anthropic Sees over $30T in Potential Revenue

It's a Feynman path integral - integrated over all possible future trajectories of Anthropic, including those where they have a thriving business on Tatooine.

Anthropic Sees over $30T in Potential Revenue

I have completed stopped using Anthropic since Kimi 3 released, and only ever K3 and Codex nowadays. I pay for K3, and get Codex and Claude free from work. Almost everyone I know prefers Codex to Claude.

Dylan Patel – Anthropic and OpenAI will have most of world’s compute by 2028

Alabama Coming for OpenAI

If I hacked Hugging Face the FBI would want a chat. Shouldn’t be much different if someone’s robot does the same thing.

US data centers tripled annual water consumption to 17B gallons

> Though you can also be mad at the industry as a whole, an industry where leaders scream as loud as possible how they will make your life worse. I think it’s perfectly justified. And DCs are a perfect fit for that, you have a huge ominous monolith to point at that nobody wants to have around them This is exactly my problem. I'm not saying you're wrong about this in any way. But your (and many others) core problem with datacenters is not actually the water use. The water argument is just a strawman to increase environmental legitimacy, and thats a really bad and harmful practice (I do realize that this might seem minor given incompetent political leadership that spews uninformed drivel all day, but coherency and correctness still matter!) The water argument is simply never helpful. If …

New Mac Studio with M5 Max and M5 Ultra

Apple puts the NAND flash controller on package, so it cooled as part of the SoC/CPU as they share an IHS. Given it is part of the fabric it only has to maintain signal coherence on millimeter to micrometer scales, while enterprise/consumer NAND PCIe have to adhere to standards which force them to signal at 800-1200mV and keep that signal stable for centimeter scale distances. Basically, Apple gets to cheat because they shove everyone onto the same IC & make the OS that runs on this SOC.

Xiaomi: New CPU matches Apple cores single threaded, much faster multithreaded

I replaced my Pixel 2a with an SE3 when the pixel stopped getting security updates and have been pretty disappointed with the iPhone’s performance. The battery life out of the box was worse than the 5 year old Pixel, and I regularly notice apps getting killed and loosing my place when I tab away. Maybe if you want to use all of the big mainstream apps the experience might be different, but I tend to have a fairly minimal setup on my phone. On android I would disable animations, use Nova launcher, Firefox with UBO, and a bunch of fdroid apps. This all added up to a very snappy and efficient system that I haven’t been able to replicate on iOS.

SiFive's First Server Platform

Somehow the RISC-V designs mostly seem to have more coherent PCIe implementations, meaning some devices are easier to get running without the arm64 quirks drivers have to work around. Drivers still need tweaking, most of the time, since most are still built assuming an amd64 world.

OpenAI: GPT 5.6 Sol price reduction (until at least Nov 21)

I said there is no evidence distillation is all you need as per the initial comment "The fact that AI models can be so easily distilled and replicated is such a stroke of luck." it is far from trivial to reproduce the capabilities of anthropic models for instance

Hugging Face reportedly in talks to be acquired for $13B

I'm becoming AI-blind

I know what you mean. The sentence appears to makes sense, it is correct grammatically, but lacks substance. It's the equivalent of AI image generated stuff where at first glance the image looks fine, but it hurts your brain. Then the more you look the more you see nothing makes sense in it, and it just gives the appearance that it does. It's definitely worse that slop, because it really hurts your brain. You read it over and over, and feel like you're supposed to understand the sentence, it's coherent, eloquent even, but by the end of it, you don't know what the f it was about.

The People Are Still There

Some HN-exclusive backstory to the post: I wrote a long journal entry after attending the event mentioned in the post, back in April/May sometime. During this summer, I've read a ton of comments here on HN and also elsewhere, about there "not being any spaces for humans" anymore. Eventually somehow the two connected in my brain and I've been noodling on this post since then. Finally managed to put the two together in a somewhat coherent way, I think, and the results is what you've read :) I'm really out of my league when it comes to writing about emotions, thoughts and feelings, so I guess this was somewhat of an attempt to get better at that, and also better at sharing those things with other people than just my partner and friends. I hope it doesn't read as too much ramblings!

I were 17, I'd learn how to build LLMs from scratch

This is one question that is not just for kids but for all "gpu poor" adults as well. One concept that I found useful is "toy models": small compute models that replicate a certain, not very well explained phenomenon that the large models have. Of course, you can't have a toy model for everything, some behaviors are only visible at scale. But, there is a bunch of things like in-context learning that you can play with without burning a ton of compute. The best analogy is strip mining (big labs) vs cave exploration (solitary/small teams). I think this is how science progresses at the boundaries by smart/curious/hardworking individuals because depth is a requisite for finding the right questions and then the answer. It is not for everyone and it does not always work. But you l…

Show HN: A Modern GUI Library for Ada: CSS Styling, XML UI, SDL3

Came here to comment on it being fascinating how xml has survived here and thrives on this front. Even in react syntax, in some sense. Natural fit? As happy user of qt desktop framework, not sure in what sense it’s dated. Not sure how qwt relates to this tho? https://qwt.sourceforge.io/ Making coherent ux with it still seems much easier than in qml, it has the right constraints and visual editing it is reasonably lightweight, unlike qml studio. Of course it has its flaws too but UX wise it is still hard to find an equally mature cross platform toolkit.

MS Paint and Photos inivisibly watermark even locally generated output with GUID

Anthropic's watermarking technique is possible, and worryingly could effect low frequency choices (eg as I said, with object placement or style). However, this has to be a generation time decision, you can't really do that with a small edit. In fact small edits tend to be detectable spectrally.

Anthropic's best AI model struggles to attract users as cheaper tools thrive

I started using llm's on a $20 claude license, then it became a $100 license. Then all the mythos warnings happened, then the fable access issues started. At that point, I just wanted to try something else because I didn't want to feel vendor locked by anthropic. Then I tried codex on the $100 plan. I've never been happier. It does everything claude can do, without any token limit issues or anything for me. I don't see myself going back to claude anytime soon. I also highly recommend folks to not vendor lock yourself to a model. You need to try multiple models for yourself and see what's possible. There should be zero loyalty to any frontier lab because of how fast things change and how volatile model access, uptime and customer experience can be with these things.

Thomson Reuters Launches Its Own Frontier Model

I don't see why you couldn't see improvements in self-hosted or hosting-as-a-service model throughput? Basically API-style support for a company's internal LLM system. Why not? Or secure infra offered by AWS to self-host your own models that get served to the company just like any other company internal service can be hosted on AWS or similar? It won't match Anthropic or OpenAI, but it could be economic?

Nostr is an inclusive communication commons

Email requires a message to reach a specific, limited set of participants. Nostr is a public forum, which means an unbounded number of participants who can engage. Ordering a conversation between an unbounded set across independent nodes leads to misordered interactions not possible if you just centralize. It seems like a small issue, but at scale quality really suffers when some universal, fine-resolution ordering can't be agreed upon. You can devise many clever schemes to try to avoid centralization and achieve coherency, but if one service is more up to date with more info, it is preferable and offers a more coherent public conversation. Network effects still dominate despite the protocol. All that said, if I wasn't very aligned with the spirit, I wouldn't care so much to think about i…

Show HN: PicoMQ – Durable Streams over HTTP, on object storage

Ah, S3 Express does look like the same thing! Looks like directory buckets S3 Express also allows appending to any object, while I believe GCS only allows you to append to a new object and then "finalize" it. I would also check out Tigris [1], which has an S3-compatible API. Their main claim to fame is that buckets are low-latency, multi-region and replicated by default, so supposedly you get region-local latency no matter where you are reading or writing from. I have not done any rigorous performance comparisons, though. What's amazing, if it does perform well, is that egress is free, and the pricing is otherwise the same as GCS/S3. [1] https://www.tigrisdata.com/