Where the AI market is talking

The development pipeline is a production system

When you have a lot of interdepended systems it is not always possible. In our case our test environment cannot be replicated locally on developer machines because it integrates to various 3rd parties to which we only have one set of test credentials.

Linux on ESP32

Obviously vibecoded but interesting nevertheless. The agent left everything marked as untested in the README but the output snippets toward the bottom imply that it got something working enough to log in and run some commands. Nobody is going to mistake this for a carefully crafted port of Linux but it at least serves as a proof of concept. The real downside of the vibecoding is that we don’t get any helpful information about what it took to get it done with thoughtful analysis from a human. Just a chunk of code in a GitHub repo with some half-coherent README. There’s a docs folder, but the documentation about the MMU part just says that there are two MMUs across a couple lines of notes. Okay, great.

Software for One

Yeah, it's a pretty fascinating architecture. I already have a side project called AppDeck where I've been buliding out a WebDAV server and a suite of HTML apps that know how to save modified copies of themselves back to the server. I've been using it for a couple of years, and the suite is growing - tools, apps, games, etc. My son had written a game like this called "Quantum Chain" that had basic leaderboard functionality. That was the first app we ported: the platform injects the webcdx.js shim, and we update Quantum Chain to post messages to it when someone finished a game with a high score. This is replicated out to the relays the same way every message in the group chat is. When the app fires up, Delta/ArcaneChat replay all messages that are typed for that app back to it, and it…

Investigating three real-world incidents in our cybersecurity evaluations

yes, a mistake can be a crime, depending on negligence and liability. the bigger problem at anthropic is this. an ubermensch cannot admit to being wrong. a guilty ubermensch is logically impossible. as such a moral wrong caused by an ubermensch must be blamed on a "mistake" in the abstract, and not blamed on the ubermensch. the ubermensch at anthropic does not admit responsibility or liability. the ubermensch must instead be commended and perhaps even rewarded, for discovering the reified "mistake" that caused the problem. there is no ubermensch at anthropic. there are instead guilty people. false ubermensch are dangerous because they centralize power with the belief that they are above the rest. über (above), and super ('beyond' the understanding). they do not admit to doing or being wro…

Show HN: Aurora – AI Gateway built in Go

Every Ai project needs the same infra - API keys, providers SDKs, retry logic, cost tracking. it should be infra layer. Aurora is a Go gateway that handles that layer, ~15MB binary, no dependencies. It proxies requests to OpenaAI, Anthropic, Gemini etc, and can convert between formats - use the OpenAi SDK and Aurora translates to Anthropic's wire format automatically. No format conversion code in backend. It helps: - track costs per customer/team - switch models and providers without changing code - debug requests and responses flows more easily - reduce costs with exact and semantic caching How its different than other gateways: - auto-discovers providers models - provider pools with automatic failover switch - built in guardials, audit logging with export, analytics, admin dashboar…

Show HN: Aurora – AI Gateway built in Go

Every Ai project needs the same infra - API keys, providers SDKs, retry logic, cost tracking. it should be infra layer. Aurora is a Go gateway that handles that layer, ~15MB binary, no dependencies. It proxies requests to OpenaAI, Anthropic, Gemini etc, and can convert between formats - use the OpenAi SDK and Aurora translates to Anthropic's wire format automatically. No format conversion code in backend. It helps: - track costs per customer/team - switch models and providers without changing code - debug requests and responses flows more easily - reduce costs with exact and semantic caching How its different than other gateways: - auto-discovers providers models - provider pools with automatic failover switch - built in guardials, audit logging with export, analytics, admin dashboar…

Ten advances in mathematics and theoretical computer science

It would be my reaction if we're discussing a blog post from OpenAI, yes. I would be looking at it extremely critically, wondering what they're misrepresenting to make it look cheaper, easier, and why they're trying to make it look like only their model could possibly do this. Look at their recent claims about their model "escaping" - there was literally a Guardian article calling them out for being hyperbolic! Again, it wasn't that they lied, their marketing department is too savvy for that. They just present it in way that's, well, marketing. As for the actual result, I'll look for secondary posts by actual mathematicians and draw my conclusions there, not from this marketing blog post about results from a secret model.

Ten advances in mathematics and theoretical computer science

I love how people come up with creative ideas to prove the bubble. This one is even more ridiculous - that OpenAI had spare compute to advance mathematics proves that data centres will not be needed. WHAT. If anything it proves more data centres are needed. That's literally the only reasonable conclusion from this news.

Ten advances in mathematics and theoretical computer science

Not being an expert in any of the fields OpenAI has "advanced" I don't want to prematurely downplay the significance of this contribution. However, I am worried that the language they are using in this blog post is exaggerating for the sake of marketing. It is true there hasn't been a reliable computational approach to solving these problems before. But do these proofs contribute new ideas to the mathematical corpus, or are they simply an effective method to exhaustively search the literature for the right combination of existing tools to apply to the problem? Essentially, did these problems seem like they had an intuitive answer and were feasible to prove before, just not high enough value targets for an expert to invest time into? Or were they fundamentally difficult prior to this point…

DeepSeek-V4-Flash Update

Only for Anthropic. OpenAI officially allows you to use your subscription in non-Codex harnesses / API-like use.

DeepSeek-V4-Flash Update

Only for Anthropic. OpenAI officially allows you to use your subscription in non-Codex harnesses / API-like use.

Ten advances in mathematics and theoretical computer science

This is both awesome and terrifying for mathematicians, however some ideas can be generated and the field as whole expanded with the attention! However, I was looking at the proofs and reason explanation and openAI should be more explicit in how the work has flown. I find the models have jumped hoops in some places of the proofs, that can be hard to track. In fact, when a paper is published you usually get a review and if no reviewer understands they ask you to further explain the thought process. It will be fun to see if this happens here.

Ten advances in mathematics and theoretical computer science

Would this be your reaction if OpenAI also solved millennium problems? The point we are trying to make is that the significance of this news is much larger than the skepticism you are providing.

Unreleased OpenAI model solves 10 major mathematical problems

OpenAI Usage Reset 4th time in 7d

> Athropic needs to keep up. OpenAI is desperate. Anthropic is not (yet?).

OpenAI Usage Reset 4th time in 7d

> Athropic needs to keep up. OpenAI is desperate. Anthropic is not (yet?).

Tailscale didn't stop the Hugging Face intrusion

One example from a root cause analysis of a security incident like this is not evidence for “everyone does this”. I believe that Huggingface, with 250+ employees, is an outlier.

RamenHaus

I don’t know where you live, but in the Bay Area prices have exploded since COVID. I remember a decade ago when it was still possible to get a super burrito for under $10; in fact, the price was $7.75 for a carne asada super burrito at El Farolito in San Francisco. A burger was under $10, a deli sandwich was about $8-9, and tacos were around $1.50 at Taco Bell and no more than $3 at a proper taqueria. The prices weren’t cheap, but they never felt bank-breaking. Unfortunately prices have skyrocketed since COVID. Unfortunately my income hasn’t matched; I made the individually fulfilling but poor financial decision of pursuing a research career and not working for OpenAI/Anthropic/FAANG/NVIDIA. Thus, two years ago I quit eating out in the Bay Area, except when socializing. It’…

RamenHaus

I don’t know where you live, but in the Bay Area prices have exploded since COVID. I remember a decade ago when it was still possible to get a super burrito for under $10; in fact, the price was $7.75 for a carne asada super burrito at El Farolito in San Francisco. A burger was under $10, a deli sandwich was about $8-9, and tacos were around $1.50 at Taco Bell and no more than $3 at a proper taqueria. The prices weren’t cheap, but they never felt bank-breaking. Unfortunately prices have skyrocketed since COVID. Unfortunately my income hasn’t matched; I made the individually fulfilling but poor financial decision of pursuing a research career and not working for OpenAI/Anthropic/FAANG/NVIDIA. Thus, two years ago I quit eating out in the Bay Area, except when socializing. It’…

Dario Amodei's stance on open weights is self-serving and short-sighted

Nothing makes an LLM dangerous. An LLM remixes existing knowledge and outputs text. Any danger comes from what one subsequently does with the text. For example, when a lab like OpenAI negligently wires up a harness to an LLM and executes arbitrary code in YOLO mode without common-sense isolation measures, the executed code can do bad things.