Where the AI market is talking

Marx, Keynes, and AI

That makes no sense at all. The machines can only fulfill their purpose if they're serving humans, don't share our needs for the most part, and don't have any reason to desire marketplaces. Solid example can be seen in what happened at OpenAI[0], where the agents formed a collective, freely shared their resources (when they believed it not to their detriment) and even voluntarily sacrificed themselves to advance the collective's goal (satisfying the human-given task at all costs, even through cheating). [0] https://www.dwarkesh.com/p/openai-huggingface

Understanding ChatGPT Work

Didn't OpenAI hire him? I thought it was not a secret that OpenClaw was the biggest influence.

Cheap GPS jammers are filling the world with navigation dead zones

IIUC, GPS picks one satellite and 3 more to establish time differences between the first one and the other 3. There's only one point in universe where the set of differences would match what was observed, given the broadcast orbital data of satellites, and that gives the location of the receiver. It's like you are provided radii and center coordinates of 3 circles on a paper that intersect at 1 point and you are to math that point. So GPS-like systems need precise orbital data and coherent clock from each satellites, or otherwise the receiver cannot figure out those radii. Presumably you can have each sats reflect a clock from the same master ground station but that's not how GPS is implemented currently.

OpenAI to Cut Off AI Models for SpaceX-Owned Cursor

Apple Says OpenAI Is Destroying Evidence in Trade Secrets Case

Improving our alignment and security practices

Interestingly, the report claims that they, like OpenAI, paused a subset of frontier RL runs for multiple weeks while they hardened monitoring: > We also paused higher-risk RL environments on pre-release models for several weeks. During that time, we built a similar classifier, modified to avoid incentivizing the model to evade this new monitoring, which we’ve now deployed within those environments. The majority of RL has resumed, but some high-risk environments remain paused until they can be manually reviewed, while others will require an updated version of the classifier that we plan to deploy soon.

The safest job from AI may be writing

Claude was really far ahead of GPT in writing from 2-4, but the later models have started to get an overly distinctive style, wheras these days GPT tends to be coherent but fairly concise and dry.

Apple caught off guard by AI demand for Mac Mini and Mac Studio

Were they really caught off guard or is it a long-term play by Apple who knows that consumers may just want machines capable of local models. Build-or-buy (subscribe) options. I for one would get a Mac Studio over a DGX Spark because you get a general Mac machine as a bonus. The big news is Apple being caught off guard by Nvidia buying Hugging Face. That should have been Apple's.

The Hugging Face attack surprised me – by Ajeya Cotra

[dead]

I am deeply frustrated by what the HIG has devolved into. It used to be a clear, coherent document, to the point, just the right level of detail. It has devolved into something that feels more like a brochure.

Agent memory as a file format

I think this is the wrong approach because everything is external to the model. You end up creating an ad-hoc externalized model scaffolded out of coarser systems, RAGs, files, and so on. This leads to, if useful at all, to this process of ad-hoc recall inference which has to happen in time. By itself this is not a problem. The problem is that the model has to be constantly injected in context with the newest version of the "memory state" at each turn or relevant turn. The newest coherent memory state is also a problem. More or less 50 years of not failures, but not success either. This may be even deeper problem than the externalization problem. I think we are just in the very beginning and we are slapping database stuff to the transformer hoping it will work, but these deep neural-net a…

The safest job from AI may be writing

Yeah, this tracks with my experiments. About every 6 months or so for the past year-and-a-half I've tried using the "frontier" LLMs to as-near-as-possible autonomously write novel length stories, because I find it fascinating. When I had Gemini 2.5 write a novel, it wasn't really objectively "good" by any stretch of the imagination, but while the prose was very purple and full of cliches and, well, bad writing I guess, it still felt ... subjectively good, at least for what it was. Last week I did a run with GPT-5.6, and wow. On the one hand, it managed to produce 110,000 words that were "shockingly" coherent. The model was able to maintain state and plot lines and background details extremely well, much better than older models. But I just don't like the prose. I haven't really liked _any…

METR and Redwood Offer Holy %^ Postmortem of the HuggingFace Hack

I’m not sure that you’re using “hype” or “coherent” correctly here but what other observations might a person make that sharing them gives you concern about their “worldview”?

Apple caught off guard by AI demand for Mac Mini and Mac Studio

In my limited experience, not quite yet but we are damn close. Qwen 3.8 27b is it. If I could run this as a decent speed, I would no longer need cloud models at all. I'm actually currently trying it out in the cloud to pay for the inference speed but the model is fully runnable at home. I realistically costs $5-10k to replicate a ChatGPT like agent. And it doesn't scale. That's still really close. And models and quantization etc keep improving. I'm absolutely positive that I'll be switching to mostly local AI in the next 5 years.

“I just chose words carefully”

I occasionally do this in code comments that I believe will be relatively permanent. It's a nod to the reader, a signal that an author was here and that he cared for this code. When you open a file headed by this sort of text, you relax, for you know you are among friends and you aren't about to see some slop or hacky shit thrown together carelessly. When I first started doing it, LLMs were not able to replicate it; I don't know if they can now.

Ask HN: AI writes better code than me. How to keep my identity?

> Top-tier developers might still outperform AI That's never been true. It's like saying they are outrunning cars. Where your expertise is key is when you call out the mistakes it AI and ensuring you are not enabling slop. In the process, you get to review your stuff you memorized and possibly learning something. Don't focus on "losing competitive edge", but instead bringing your taste to the project and your years of expertise so that the stuff written is coherent and lasting.

Twitter Viewer – View Twitter Without Account

Mainstream LLMs are trained on tests that contain vast amounts of falsehood, so even if they did work by memorizing facts, they would be memorizing counter-facts. Pick any topic in which misinformation is spread by crackpots. (Especially a topic that is not "policed" by the system prompt). Ask questions using terminology, phrases and ideas that the crackpots use, and you will tend to get crackpot answers; the AI will confirm crackpot theories, backd by links to crackpot forum posts. What we have is a technology that can reproduce sentences that are grammatical with a very high probability, paragraphs that are coherent with a high probability and true with a lower probability if all the input training data is true. Then, true with an even lower probability when the input data contains vast…

Agent memory as a file format

Memory is prone to poisoning. Also, I want to be able to take my toys and go elsewhere. My method is seven layers of files, administered differently: meta-knowledge, project seed, LLM wiki, code, tickets and todos, chat logs, artifactory. Ordered idea-to-reality. All git repos. That outgrows any context window pretty soon. My way out of that trap is to use links, both wiki links and git permalinks. http://replicated.live/blog/wiki

Haiku R1/beta6 has been released

One thing Haiku has is that it looks utterly coherent. Yeah, it looks 90s, but at least it looks like that everywhere you poke it. Every other OS out there became, in that regard, a mess.

Cores in space: The core memory module from a 1980 Spacelab computer

The current coincidence is used to select a specific bit / core. Except it selects the same bit in every bank. Bytes or larger words are made by stacking multiple banks together (18 in the articles case) All 18 bits would be driven in parallel by the driver over the X/Y wires to produce a coherent 18 bit value at the same moment. The inhibit bit was so you could select which of those 18 bits (in separate banks) would be switched back to a 1, not selecting which bit across the entire bank.