Where the AI market is talking

Bun 1.4 Rust rewrite is not looking good?

Lots of people are apparently happily using the canary version in prod. Anthropic has been using it for ages now, as has Prisma. Why not release 1.4 (or 2.0 or whatever) then? It won't get picked up by lots of tooling and won't get allowed in many "no pre-release in prod" shops until then. It had plenty of issues before (many thousands of open issues) so a few thousand more is hardly a big deal...

OpenRouter is joining Stripe

I don't know about cheaper. The flexibility is kind of what I'm pointing to, if I understand you right. Instead of dealing with Anthropic, OpenAI, XAi etc directly and having systems to handle user assignment, budgeting etc for each, companies just deal with GitHub, which they've probably been dealing with anyway.

Feature Request: Support AGENTS.md

Doing the whole “pay per token rate” when logging into a harness not made by them is not only being enforced by anthropic/claude though. It also is what Google does with agy (their TUI). Is there any provider that doesn’t do this? That’s the one I want to support. > TLDR: they are all probably doing it because they are banking on you not using all your tokens. I would be surprised though, it makes business sense to make the default vendor native TUI cheaper because it’s understood that most nascent users will just use the harness offered by the LLM provider and those that would stray off that would probably be more power users who would spend their token share more consistently till it drains.

Feature Request: Support AGENTS.md

Shoot, yes. The user experience of logging in is the same, but yes- using a different (non-cc) harness bumps you up to the pay-per-token rates. However! I have seen projects which use CC under the hood, in order to get the subsidized rates. Writing comments which get me to go look through anthropic documentation, and find friction I wasnt aware of (CC does not have an app-server), refreshes my frustration towards anthropic.

Round Hill Music Sues Suno and Anthropic for $1B over AI Training Data

Feature Request: Support AGENTS.md

It’s easy to end up with bunch of old stuff in AGENTS.md that latest models no longer need and what might just confuse them. See for example the story about Anthropic cutting 80% of their system prompt [1]. Maybe they feel just using the AGENTS.md written for another model and possibly different era gives bad user experience. [1] https://claude.com/blog/the-new-rules-of-context-engineering...

Feature Request: Support AGENTS.md

They did walk that back, you are able to use an Anthropic login in a non-anthropic harness. (For now.) (Edit: I failed too mention, using a different harness with a CC/Anthropic subscription bumps you up to the pay-per-token rate.)

Feature Request: Support AGENTS.md

Oh my god, that's an official plugin. That's hilarious. I had the same idea last year, when I noticed how much cheaper GPT was for the same tasks, but Claude was still better as a high level "operator". Also cause OpenAI was cool with you calling their sub in an automated way, but Anthropic very much was not (they were banning people at the time).

AI Hurtles Ahead

it drones on and on, it lacks a coherent train of thought I assume it is agent written, slop regardless I'd be very hesitant to call a piece of writing about Ai "watertight" when the author starts with > I want to start off with my usual caveats: I’m not active in the stock market; I merely watch it as the best barometer of investor psychology. I’m also no techie, and I don’t know any more about AI than most generalist investors. But I’ll do my best. Sampling the author's priors also shows low quality imo

Feature Request: Support AGENTS.md

You're asking the wrong question here. There is: How exactly does Claude Code* not support AGENTS.md? How exactly does the Claude suite of Models not support AGENTS.md? The problem being pointed at in parent linked is referring to the Claude Code Harness not supporting AGENTS.md. Harnesses which do support it (eg codex), append the content to the initial model turn upon the model discovering it at the project root. * Or Claude Code Tui, Claude Code desktop, Claude Cowork, Claude Desktop, Claude Design, Etc. (Edit: i wish I could tattoo the distinction on ny forehead. Im vocal about Anthropic engineer-oriented tools being lackluster, and i frequently find myself in conversations where I have to stop a coworker and ask if they're talking about a platform, tool, or model- and which, dependin…

Feature Request: Support AGENTS.md

Stop supporting Anthropic with your money. They are a hostile company.

Bun 1.4 Rust rewrite is not looking good?

Both of these are true. I recall Anthropic's lofty claims about their C compiler. It just did not work, and they claimed all of this and that. It was clear they were driven to believe that agents alone could write a C compiler, and refused to believe they couldn't. Not that I think that AI couldn't write a C compiler. It definitely can... with a human at the helm and sufficient time. But, for people who don't like AI, they also refuse to believe that AI could be good at any particular workflow. Which is just obviously not true, there are workflows that AI is really good at. Language re-writes are one of them. They're almost mechanical in nature and trivial to check via tests. So, the perfect use-case for an LLM. I'm not saying the Bun re-write is any good, but the underlying concept is so…

In South Korea, some savour last season of legal dog meat

I get to coherence by concluding that it would be ethical to eat my dog. I just prefer not to.

Router by Ramp

Muse Spark, Gemini, mistral and something from cohere would be nice That said the current models ain't bad!

In South Korea, some savour last season of legal dog meat

If you were actually looking for coherence you would consider that both are unethical. The only reason you’re not actually adhering to coherence, as you said, is your fondness for eating cows and pigs. Thats hardly a justification for an unethical act.

Chain-of-Thought Reasoning in the Wild Is Not Always Faithful (2025)

It is and has been - skipping the tokens causes performance to drop. But replacing the tokens with filler causes the performance to drop, but by a lot less. You can also train models to emit broken or unrelated thinking tokens and their performance is also not much worse than the ones that are trained to output somewhat coherent thinking traces. This points to a hypothesis that the content of the tokens is only slightly related to the mechanism by which it improves performance, and that primarily the extra tokens allow the original prompt to be processed more deeply by the model, because earlier tokens will essentially pass through the model many more times than later ones.

Opus 5.0 drives incoherence into the stratosphere

Your description grants psychological characteristics like intelligence, will, and intent to the AI. The AI is an intelligent and willful agent that has some intent to take over the firm and is doing that by making strategic hiring decisions. A different and externally/phenomenologically identical way to look at it is that the AI is just a program that generates output too unpredictable, too voluminous, or too idiosyncratic for people to evaluate. When people submit their own will and their own intent to that program, by treating the black box as an intelligent oracle, they've entered into a psychotic state divorced from reality. It's two self-consistent and coherent perspectives of the same event, except one involves believing that scifi AI has arrived and the other just thinks peop…

Mathematics in the age of AI

What's wrong with sitting on the beach with a bottle of wine for eternity, with no need to do anything, knowing that all your needs and desires will be automatically taken care of? I don't think you can be coherently pro-AI without thinking that the human brain will be obsolete, unless you believe in some inherent magic that the brain is imbued with. The only other option is that you haven't thought through the long term consequences of the innovation.

OpenRouter is joining Stripe

To add to this; payment infrastructure requires a lot of heavy lifting. There's a lot of regulations you need to adhere to, different payment systems in different countries, settlement, chargebacks, etc. There is a reason why doing your own payment processing is a thing. OpenRouter may have some interesting things in streamlining the process of switching LLM providers, but it is indeed something easy to replicate in comparison to payment processing.

Opus 5.0 drives incoherence into the stratosphere

According to Claude....When asked to explain some obvious biases and its tendency to adhere to established heresay over novel information... it told me it judges veracity and "quality" of facts based on: "statements against interest outweigh costless ones, replicated/checked claims outweigh untested ones, large logged samples outweigh single memories — not because memories are worthless, but because those methods close off specific, known failure modes a memory alone can’t. Could I be running that standard with unconscious asymmetry, harder on the anti-Claude data than the pro-Claude data? Genuinely can’t rule it out. The fix for that is holding the standard explicit and checking it against itself "