mohsen1

Proofline | Founding Staff / Principal Engineer | REMOTE (Worldwide, Berlin timezone overlap preferred) | Full-time | €120k-€200k + equity | https://proofline.sh/careers/founding-staff-principal-engine...

Hi HN, I'm the founder of Proofline. We're building an infrastructure-aware AI code reviewer that catches deployment failures before they reach production. Think of it as a type checker for your cloud infrastructure. We analyze code changes against Terraform, live cloud environments, and deployment evidence rather than just reading diffs.

We're hiring a hands-on Staff / Principal Engineer to help us build. You'll own core SaaS architecture, cloud integrations, developer tools, and product delivery, with a path toward engineering leadership.

Stack: Rust, Go, PostgreSQL, React/TypeScript, Terraform, AWS, GCP.

We're looking for someone with 10+ years of hands-on infrastructure engineering experience, deep SaaS and cloud expertise, extensive use of AI coding agents, and a track record of shipping end to end.

Remote globally, flexible hours, regular Berlin visits required. Salary: €120k-€200k depending on experience and location, plus meaningful founding equity. No visa sponsorship.

To apply, ship a small feature PR to one of our open-source developer tools and email [email protected] with the link. If you have 10+ years of relevant experience and an established online presence showcasing your technical work, you can skip the PR and share your best work with us instead.

Details: https://proofline.sh/careers/founding-staff-principal-engine...

to be fair `gh` works so much better for me than crafting curl requests (even with LLMs)

I ran out of patience before models got good enough to get this done.

https://tsz.dev

Congrats to the team on getting this out

Since it is fast and understand images, I wonder if it can play video games. I have a harness setup for the LLM play EA FC but even the fastest LLMs are too slow for it. I need to try this with Decisions API

> agents were not told to 'go hack'

I was referring to the HuggingFace incident.

> none of the alignment techniques that are applied to models today actually work

none of techniques to autonomously drive a car was/is not working for a long time. no company came out and said 'this is impossible to do, let's change the regulations'.

I listened to Jensen Huang's interview with Ezra Klien and it was so refreshing to hear it from an engineer. Jensen framed it as OpenAI's responsibility and recklessness which I agree with. Jensen thinks it's an engineering problem to build better sandboxes.

It's irresponsible for OpenAI to give unaligned agents a prompt to 'go hack' and internet access. They know better, so I am thinking they might have other intentions to let those swarms have any sort of internet access.

"100 grams" is the headline here

Why tho? AI Studio has this model available and I don't need to paste the API key anywhere

https://aistudio.google.com/generate-speech?model=gemini-3.8...

> The SAME model may even give different answers to the same prompt when asked multiple times

temperature?

see sister comment's response https://news.ycombinator.com/item?id=49786548

This is less true for modern posttrained models. Model identity can be explicitly reinforced during posttraining. Qwen's own finetuning docs include identity training examples, and Qwen models have been trained with system prompts that explicitly say things like "You are Qwen, created by Alibaba Cloud."

So a model correctly identifying its family doesn't necessarily mean it inferred that from pretraining.

I think with Jev, they took a posttrained model and trained it further, so it did not forget about its earlier knowledge during Owen's own RL.

I can't find it but saw that if you give Jev English alphabet as choices and ask it in a loop what model it is, it would say Qwen

also tried myself: https://console.typesafe.ai/playground?share=shr_1690a3160f1...

There is an open PR for VLLM to do this via DefussionGemma

https://github.com/vllm-project/vllm/pull/57250

yup https://github.com/vllm-project/vllm/pull/57250

> A little research before vibe-coding an entire language and compiler could have substantially improved the result because the author would have known what to ask for.

A little research before writing and publishing a personal attack like this could have substantially improved the result because the author would have known what they're writing about

Victor is not a formal verification noob as this article suggests

if you were disappointed by iOS dictation, give it another shot after this update. It has improved a lot. I've been using it a lot more than before

I speculating but hard to not see that DeepSeek is brewing a full Pro model with those new techniques to come out right around the time of Anthropic and/or OpenAI IPO to tamper the excitement for their offering.

Turkish as well. They moved from Arabic to Latin script.

I absolutely loathe my LG smart TV. Just switching to another input requires 3 button press.

I hope someone uses their free time to hack the OS and offer a clean and simple interface to make it a “dumb” TV.

I have a question about “domain expertise” as a component of future knowledge worker requirements. How does one gain such expertise in a context where thinking is expected to be delegated to AI (shifting from problem solving to question asking, as noted in this paper)?

How does one learn to pose the right questions when basic ones are rarely “manually” answered? that is, without an AI assistant’s help

This pattern appears in schools, where AI interferes with human development that typically demands long and difficult effort of actually answering questions

> The company on Wednesday confirmed speculation that the Ox Alpha model is a new iteration of its GLM series and said it will release the weights for it tonight, in response to queries by Bloomberg News.

Seems legit.

It's really hard to know how good it is. So much hype around it.

wow i totally missed "It Was Just an Accident". Have to watch asap.

Hmmm... I had a different experience. They had a fully automated environment where you ha to write test that passes some tests. No human involved. And the time requirement was insanely tight

That's a lot of words to say you need larger set of questions for today's models. 300 questions won't be enough to find the difference

> one of the basic tenants of algorithm development was that you can't just brute-force your way towards a solution for some complex problems

Mote-Carlo is pretty useful still. Not sure if your statement holds

Do you think any programmer really understands how their program works end-to-end? At some abstraction layer, we're all clueless. There are many layers between what you type into the text editor and the actual CPU ticks that make your program work. I bet nobody fully understands the whole stack.

Now that that text editor accepts English, we're all calling each other names, etc.

RL. Lots of RL