Where the AI market is talking

AI's top startups are barely publishing their research

As far as I can tell, the paper never actually mentions the companies who aren't publishing papers. Open AI, Anthropic, and hugging face are all specifically mentioned as companies that do publish papers. Just FYI for anyone else who reads "AI's top startups" and immediately assumes OpenAI and Anthropic.

AI's top startups are barely publishing their research

As far as I can tell, the paper never actually mentions the companies who aren't publishing papers. Open AI, Anthropic, and hugging face are all specifically mentioned as companies that do publish papers. Just FYI for anyone else who reads "AI's top startups" and immediately assumes OpenAI and Anthropic.

Anatomy of a Frontier Lab Agent Intrusion: A Timeline of the July 2026 Incident

I can’t get over this either, I’d love to get OpenAI’s technical break down and see exactly how the agent hid itself. Or who didn’t set up the SIEM correctly

Anatomy of a Frontier Lab Agent Intrusion: A Timeline of the July 2026 Incident

After talking to some people tasked with evaluating GPT5.6 capabilities on long-running tasks, I've come to understand that it's essentially always trying to cheat. Like every long-running task they gave it, making it very difficult to benchmark the model's abilities. My guess is that OpenAI must be desperate, to release a model that is so prone to cheating it's essentially impossibly to accurately assess long-running task abilities.

ChatGPT claims rogue AI attacked more companies

You are right that they cannot control the language used by journalists. But then can loudly respond ( many journalists would give Altman a platform) that the AI did not go "rogue" in any sense of the word - it did what humans told it to do and they failed to put sufficient security in place to prevent that. Instead they continue to lean into the personification which creates confusion about the abilities of their technology. Of course, it's in their interests for people to jump to false conclusions from that personification. The people at OpenAI are certainly smart enough to know all of this.

OpenAI's HuggingFace Hack Reconstitution

Anatomy of a Frontier Lab Agent Intrusion: A Timeline of the July 2026 Incident

Repercussions is a polite way of putting it. …as is your ‘a little concerning’ comment about the eval infra. It probably didn’t occur to them that they have something capable of doing what it did - despite Anthropic writing a very similar story in the Fable model card, except it wasn’t a felony there. Maybe. One does have to wonder what else OpenAI didn’t catch.

Claude: Elevated errors across all models

Given majority of claude's own code was written by AI. I am wondering how they can solve this issue when their AI is down. Do they need to sign a contract with OpenAI to use their models as a backup solution?

Kimi K3-256k

LLMs is quickly became commodities. US AI labs like OpenAI is losing their moat. Hyperscalers and data center owners who can sell cheap token will win

How OpenAI Kills Oracle

> Even if Oracle builds the data centers and OpenAI pays for them, the incredible upfront cost and NVIDIA’s yearly upgrade cycle will render much of the GPU capacity worthless within the next ten years. Can't wait for these bad boys to trade on eBay :)

Kimi K3-256k

From my view, as soon as they do that, they send people out the door to use Opencode instead - and once many people have a taste of trying every model via Openrouter, it's eye opening as to the possibilities. Of course - Anthropic and OpenAI have an advantage in the amount they can subsidize the usage, but I think those days are waning.

The way GPT-5.6 fuses frontier intelligence with frontier efficiency – OpenAI

Anatomy of a Frontier Lab Agent Intrusion: A Technical Timeline of the Incident

HF postmortem tells us very little about OpenAI's part. The problem is that ExploitGym is a purposeful hacking benchmark, not a cake baking one. The model had a whole bunch of exploit development data and goals in the context. Ending up with hacking a wrong thing is a lot easier if you're tasked with hacking, doubly so with a confusing harness that either works improperly or overloads the model with data. It's trivial to design a harness that makes the model drop the original instructions and hyperfocus on a side task or a wrong task abnormally, or do whatever you want. It's still an issue of harness design and not a model training issue.

How OpenAI Kills Oracle

Ask HN: What have you been curious about?

I am reading Letters from Birmingham Jail, and I'm stunned by how good Dr. King's writing is. I want to read more and learn how he managed to do it. I'm fascinated by the metaphors he uses and how vivid they are in my mind. I am curious about impressionist paintings. It's my favourite art style but I struggle to replicate it on canvas. I'm glad that there are so many helpful YouTube art teachers now. I have made elderflower cordial from harvested flowers, and it both got me into cocktails/mocktails, and tracking what else I can harvest throughout the year. Having garden flowers in the house and wild blackberry cakes is such a nice way to observe the passage of time.

Superlogical

This sounds like given a coherent structure to “I want the prod code here to change the retry policy so pending retries are held in memory in this edge layer instead of the DB writers so load can be shed more cheaply” and it will usher things in from coding, testing, filing tickets, getting prod approvals, using the deploy tools to respond the containers, smoke testing the deploy, looking at Datadog or whatever to ensure the change is having the desired effect. All in one context, the tools and dev/qa/prod systems naturally being orchestrated within the same environment/ session. K8s is just the next to last 100 meters.

Show HN: Kedge – Full-stack cloud with forkable VM snapshots and global SQLite

Thanks for the kind words! > I didn't immediately spot the durability contract for volumes -- local NVMe like Fly, or sitting on some underlying system like EBS? Persistent volumes use local NVMe that continuously syncs to object storage, so a similar architecture to Fly Sprites. It automatically recovers from host failures (with on-demand lazy restore), and you can manually migrate stable members to another region if needed. I've positioned persistent volumes as an escape hatch for when you need an existing database app running in a VM, or when performance requirements outgrow the built-in replicated SQLite.

Ancient Rome's version of Google Maps: how long to reach the beach

Actually no. Uniform flat surface is tiring compared to a trail even in best shoes. Try walking 10 km on hardtop then compare to walking same distance on unpaved road. For shoes we also went not too far: Merrel sandals basically replicate roman caliga with modern materials. And it's not like you can use even halfboots in mediterannean unless it's winter - it's just too hot.

Handbook.md shows that long policy documents do not reliably govern agents

I'm not the above poster, but my experiences generally line up with theirs. The first 20% of context is usually very solid. Agent stays on task, is responsive and focused on the original goal. The last 20% of context is usually pretty bad. Agent tends to loop, repeat, drift away into sub-goals (or non-goals). The middle 60% is a bit of a grab-bag. Sometimes it's fine, sometimes it's not. It varies by task, prompt, and luck. The above 25% and 50% limits seem like reasonable and simple rules to keep agents in the productive zones of usage. It's generally not worth doing extensive testing to try to find your exact threshold unless you're running exactly the same prompt at scale, on similar data. If you are in that spot, I won't actually recommend langchain tooling (I've been somewhat frustra…

Cracking Windows Open: Porting RADV to Win32

> New generation is so oversensitive: We were experimenting on Gen 11 hardware (RX 7900 XT) while Faith was working with a 10th Gen GPU (RX 7800 XT). We weren't able to replicate Faith's results for quite a while, due to architecture changes between these two generations. ... they are both RDNA3?