Where the AI market is talking

Anthropic's ‘watermark’ text adulteration in Claude is a perversion of writing

> “By definition it must make text worse … because the nature of the watermarking algorithm requires it to sometimes increase the probability of selecting a worse word choice and decrease the probability of selecting the model’s best choice.” Gruber made an effort to but doesn't fully understand how SynthID works. LLMs select the next word randomly from a set probability distribution, so there is no "best choice" unless you run the LLM with a temperature of 0 which would give out terrible results. Anthropic runs a non-distorting version of SynthID that doesn't change the probabilities of the underlying distribution of tokens. It makes the watermark less likely to work over smaller samples but preserves text quality. I encourage the mathematically inclined to read the paper: https:&#x2F…

OpenAI President Greg Brockman: "You Need to Believe"

Qwen3.8 27B scores 52 on Artificial Analysis

i think there definitely is some truth to this in terms of embeddings spaces, which is why i believe they are implemented by OpenAI/Anthropic in roughly highest import => least import bit order-- an overwhelming majority of the variance is in the first few hundred vector bits. i haven't actually tested this myself by manually truncating vectors, but it is my understanding that they generally speaking have this property.

Qwen3.8 27B scores 52 on Artificial Analysis

i think there definitely is some truth to this in terms of embeddings spaces, which is why i believe they are implemented by OpenAI/Anthropic in roughly highest import => least import bit order-- an overwhelming majority of the variance is in the first few hundred vector bits. i haven't actually tested this myself by manually truncating vectors, but it is my understanding that they generally speaking have this property.

Launch HN: Speko (YC S26) – OpenRouter for Voice AI

Gemini 3 flash series is quite good, also elevenlabs scribe v2

Anthropic's ‘watermark’ text adulteration in Claude is a perversion of writing

While Anthropic is sharing this publicly, there’s isn’t much reason that other models could quietly be doing this or start. Local models could probably catch some patterns.

On AI regulation and messaging

D.A. >> Anthropic is ramping up its efforts very quickly in biology and medicine At the same time, the best Anthro models refuse to answer biology questions, keeping the public in the dark not only about new AI-assisted results but also about conventional and health-protective research. D.A. >> we hope to have incredible results in the coming years and some early glimmers in the coming months. So, what changed? He went from a promise of AI bonanza any minute now to a promise to maybe cure cancer in the coming years , that's right, an unspecified number of years . Until that time, we're supposed to eat glimmers of hope because the AI-caused inflation will make food unaffordable. D.A. >> When we’ve actually accomplished something real, the whole world will hear about it, a…

Anthropic's ‘watermark’ text adulteration in Claude is a perversion of writing

I agree with you as long as things are at a small scale. But at a large scale, these things reshape society and what it means to be human. Social networks started out[1] as almost wholly good. The negative effects came from scale. Same with advertisement-funded websites and tons of other things that started out as being overall positive for the commons and ended up being highly negative. You can't just stick your fingers in your ears and ignore what is happening in reality. > Anthropic is crowing about this achievement because they are afraid of the dirt cheap AI models coming out of China and eventually other places impacting their valuation. Full stop. It's the "full stop" that I'm disagreeing with. Yes, you can make valid arguments about the motivations behind this, or the effective…

Memory prices climb 500% in 12 months, up to 10x the lowest ever tracked prices

There also seems to have been a major shift in strategy from AMD / Intel / Apple / NVidia due to this, as I was used to CPU / GPU / chipset upgrades coming out more regularly with double-digit percentage improvements each year, but that seems to be basically over for the consumer space now that everyone is just trying to build as much AI hardware as fast as possible for enterprise markets. I also used to rebuild my desktop every 1-2 years and upgrade laptops every 2-3 years, but I'm currently a bit past that and have no plans to refresh anything in the near future. I'll probably wait another CPU and GPU generation or two to even consider it, especially with prices being this ridiculous. I'm fully anticipating AI hardware demand to dry up once Anthropic and OpenAI …

Memory prices climb 500% in 12 months, up to 10x the lowest ever tracked prices

There also seems to have been a major shift in strategy from AMD / Intel / Apple / NVidia due to this, as I was used to CPU / GPU / chipset upgrades coming out more regularly with double-digit percentage improvements each year, but that seems to be basically over for the consumer space now that everyone is just trying to build as much AI hardware as fast as possible for enterprise markets. I also used to rebuild my desktop every 1-2 years and upgrade laptops every 2-3 years, but I'm currently a bit past that and have no plans to refresh anything in the near future. I'll probably wait another CPU and GPU generation or two to even consider it, especially with prices being this ridiculous. I'm fully anticipating AI hardware demand to dry up once Anthropic and OpenAI …

On AI regulation and messaging

If Anthropic is the first to develop AGI, then everyone else becomes subordinate to Dario Amodei? And if DeepSeek is the first to get there, everyone instead becomes subordinate to Liang Wenfeng? I cannot believe that there is a threshold of intelligence where getting there first means a decisive and durable advantage. If you do believe that getting there first means ruling the world, then our viewpoints are probably too far apart for mutual understanding.

Memory prices climb 500% in 12 months, up to 10x the lowest ever tracked prices

It's pulling the ladder up behind them. OpenAI bought up more co tracts for wafers than they could even use. But now that means everyone else has to fight over what's left. This makes it too expensive to compete with OpenAI. No upstart can really afford to build infrastructure.

Memory prices climb 500% in 12 months, up to 10x the lowest ever tracked prices

OpenAI (Stargate) kicked off the price increases a year ago by locking up 40% of the global supply of RAM output. https://www.tomshardware.com/pc-components/dram/openais-star...

Anthropic's ‘watermark’ text adulteration in Claude is a perversion of writing

These are services, so "how would <x> do this today?" is irrelevant. The real question is: "What can manipulation of pseudo-random number generation do?" We know that in the cryptographic world, attacking "randomness" is a key offensive capability. It will be here as well -- if Anthropic can watermark text as generated it's LLM, will it be able to watermark outputs as generated by "Spooky23/FooCorp"? Can I pay Anthropic to steer inquiries in a way that benefits my company or governemnt? Pseudo-random to the end user appears random. Most treat it like a random chance. It is not.

Anthropic's ‘watermark’ text adulteration in Claude is a perversion of writing

So Anthropic goes on the shit list with Sony Blu-ray and its cinavia watermarking.

Fossil – A Coherent Software Configuration Management System

Anthropic's ‘watermark’ text adulteration in Claude is a perversion of writing

> So I agree that the watermarking has a cost. But you can't leave out that it is an attempt to reduce negative externalities of AI. Whether it's a realistic or worthwhile attempt is a whole other debate (and Gruber does a good job of debating just that in the latter part of the essay), but saying that the user's needs are the only thing that should ever be considered is reprehensible. Computers are tools that exist to serve. Creating some bizarro process where we are compromising the technology in service of it's owner to achieve some nebulous goal is gross. Anthropic is crowing about this achievement because they are afraid of the dirt cheap AI models coming out of China and eventually other places impacting their valuation. Full stop. There's some vague notion of preventing harm wit…

Training AI Scientists to Replicate Research

Launch HN: Speko (YC S26) – OpenRouter for Voice AI

Ever since OpenAI launched their improved voice mode, I've been looking for a capable LLM with builtin voice-in and -out. Llama 4 was supposed to be it but turned out to be a dud. I haven't followed the topic closely lately, did I miss anything? Are there capable (!) open weights omni models that allow low latency voice chat? If so, what software do you use them with? Can you use a PWA on your phone? WebRTC? WebTransport?

Incident with Github.com

Yeah let's not forget that in those brief few days when Sam Altman was fired, Satya Nadella swooped in and hired him. Despite the failure of that move, Microsoft has invested $13 billion directly into OpenAI, and has spent $100 billion on the partnership. No sympathy from me watching their software crumble as a result.