rdli

I would also say: I think this works great because the goal is well-defined and measurable.

I’ve also used Opus 5.5 on some hill-climbing, and a lot more steering is required here, because … eval is hard.

I’m on the $100/month subscription; this session took about $500 in token-equivalent costs.

(Note that it wasn’t all Opus 5.5; I have a setup that uses Fable 5.1 as an advisor, Sonnet 5.5 for mechanical changes, etc.)

Yes. It spawned multiple subagents to run different experiments to benchmark a lot of different things, reviewed CI logs from past runs, etc. In the end, there were changes to what/how we cached, various code quality checks, speeding up test runners, and many other things.

It’s a really good model. Over the past few days, I give Opus some general directives to basically speed up our CI, and telling it I care both about billing minutes and wall clock time. I told it to create a plan after analyzing everything in our CI, run the plan by a Fable subagent, and then focus on low-risk, high-reward changes.

9 hours later, I had 12 PRs ready to be merged, and the net result is CI time has dropped from ~10 minutes to ~4 minutes, and billing minutes have dropped around 60%. Less than an hour of my attention.

I couldn't find a client of any sort for WeatherNext, so I had Claude write a simple one: https://github.com/richarddli/weatherodds.

One thing that wasn't obvious in the post: it's usage (token) based pricing, not including in your existing Claude subscriptions.

Cybersecurity/AI seed startup | Founding AI Engineer | Bay Area | Full-time

We're seed-staged, 3 people, building an AI for cybersecurity, looking for a founding AI engineer who wants to learn/apply SOTA techniques for AI. Ideal background is experience building production agentic AI systems (by this I mean something like Simon's definition: https://simonwillison.net/2025/Sep/18/agents/ definition) who also likes to think about WHAT to build and not just how.

We are a Golang/Python shop (although I'm not sure that matters so much any more).

Email jobs@polarsky.ai with subject HN.

I get that but just not entirely obvious how you do that for the Notion AI.

Securing LLMs is just structurally different. The attack space is "the entirety of the human written language" which is effectively infinite. Wrapping your head around this is something we're only now starting to appreciate.

In general, treating LLM outputs (no matter where) as untrusted, and ensuring classic cybersecurity guardrails (sandboxing, data permissioning, logging) is the current SOTA on mitigation. It'll be interesting to see how approaches evolve as we figure out more.

Nothing super-fancy. We have a common GitHub repo in our org for skills, and everyone checks out the repo into their preferred setup locally.

(To clarify, I meant that some engineers mostly use CC while others mostly use Codex, as opposed to engineers using both at the same time.)

This is great. At my startup, we have a mix of Codex/CC users so having a common set of skills we can all use for building is exciting.

It’s also interesting to see how instead of a plan mode like CC, Codex is implementing planning as a skill.

Polar Sky | Bay Area | Full-time | Founding AI Engineer

Generative AI is rewriting how organizations use data, and breaking traditional security models in the process. We’re a team of cybersecurity, AI, and systems experts building the foundation for secure, trustworthy AI in the enterprise.

We're looking for a Founding AI Engineer who loves building with AI -- crafting context pipelines, integrating and evaluating LLMs into production systems, and delivering AI-native product experiences. You'll work on all parts of Polar Sky, from the data and eval systems to the reasoning, retrieval, and orchestration systems.

Apply online here: https://ats.rippling.com/polar-sky/jobs/a04ed5b7-6202-45e6-b....

Polar Sky | Founding AI Lead | Bay Area/Seattle | Hybrid/Onsite | Full-time

We're a well-funded, pre-seed cybersecurity startup focused on data security. I'm looking for a founding AI lead with experience in fine-tuning LLMs (expertise around RL + reasoning models a big plus). This person would own the full AI stack from data to training to eval to test-time compute.

Who's a good fit:

* If you've always thought about starting a company, but for whatever reason (funding, life, idea), this is a great opportunity to be part of the founding team. We're 2 people right now.

* You enjoy understanding customer problems and their use cases, and then figuring out the best solution (sometimes technical, sometimes not) to their problems.

* You want to help figure out what a company looks like in this AI era.

* You enjoy teaching and sharing knowledge.

Questions, interest, just email jobs@polarsky.ai.

Polar Sky | Founding AI Lead | Bay Area/Seattle | Hybrid/Onsite | Full-time

We're a well-funded, pre-seed cybersecurity startup focused on data security. I'm looking for a founding AI lead with experience in fine-tuning LLMs (expertise around RL + reasoning models a big plus). This person would own the full AI stack from data to training to eval to test-time compute.

Who's a good fit:

* If you've always thought about starting a company, but for whatever reason (funding, life, idea), this is a great opportunity to be part of the founding team. We're 2 people right now.

* You enjoy understanding customer problems and their use cases, and then figuring out the best solution (sometimes technical, sometimes not) to their problems.

* You want to help figure out what a company looks like in this AI era.

* You enjoy teaching and sharing knowledge.

Questions, interest, just email jobs@polarsky.ai.

Seems that OpenAI is acquiring Io for $6.4B in an all-equity deal.

I would think that that the NVidia Dynamo SDK (pipelines) is a big difference as well (https://github.com/ai-dynamo/dynamo/tree/main/deploy/sdk/doc...), or am I missing something?

In this analogy, Dynamo is most definitely not like Django. It includes inference aware routing, KV caching, etc. -- all the stuff you would need to run a modern SOTA inference stack.

This is really interesting. For SOTA inference systems, I've seen two general approaches:

* The "stack-centric" approach such as vLLM production stack, AIBrix, etc. These set up an entire inference stack for you including KV cache, routing, etc.

* The "pipeline-centric" approach such as NVidia Dynamo, Ray, BentoML. These give you more of an SDK so you can define inference pipelines that you can then deploy on your specific hardware.

It seems like LLM-d is the former. Is that right? What prompted you to go down that direction, instead of the direction of Dynamo?

The blog post was a little unclear, so my summary was:

- They used QwQ to generate training data (with some cleanup using GPT-4o-mini)

- The training data was then used to FT Qwen2.5-32B-Instruct (non-reasoning model)

- Result was that Sky-T1 performs slightly worse than QwQ but much better than Qwen2.5 on reasoning tasks

There are a few dismissive comments here but I actually think this is pretty interesting as it shows how you can FT a foundation model to do better at reasoning.

I took a brief look (~5 minutes). My $0.02 is that it's not clear what problem you're trying to solve. I get what some of the features do (e.g., templated prompts) but it would be v helpful to have an example of how you actually use magentic, versus the non-magentic way. It feels like a lot of syntactic sugar, if I'm being honest (not a bad thing, but something you might want to be clear about, if that's the case.)

(author here) I didn't put this in my post, but one of my favorite moments was when I read some of the LlamaIndex source code which pointed to the GitHub commit where they copied the code verbatim from LangChain. (LangChain is MIT-licensed, so it's OK, but I still thought it was funny!)

Not a bad move by Red Hat. Red Hat lost the battle of the cloud to Azure, AWS, and Google, but AI is still a nascent space. vLLM's deployment model fits neatly into Red Hat's traditional on-premise / support-centric business model.

I'm running ollama, but it's still slow (it's actually quite fast on my M2). My working theory is that with standard cloud VMs, memory <-> CPU bandwidth is an issue. I'm looking into vLLM.

And as to sidestepping inference, I can totally do that. But I think it's so much better to be able to ask the LLM a question, run a vector similarity search to pull relevant content, and then have the LLM summarize this all in a way that answers my question.

Thanks! I should have been clearer -- embeddings are pretty fast (relatively) -- it's inference that's slow (I'm at 5 tokens/second on AKS).

I'm working on something like this! It's simple in concept, but there are lots of fiddly bits. A big one is performance (at least, without spending $$$$$ on GPUs.) I haven't found that much in terms of how to tune/deploy LLMs on commodity cloud hardware, which is what I'm trying this out on.

This is cool! I've been trying out bits & pieces of the RAG ecosystem, too, exploring this space.

Here's a question for this crowd: Do we see domain/personalized RAG as the future of search? In other words, instead of Google, you go to your own personal LLM, which has indexed all of the content you care about (whether it's everything from HN, or an extra informative blog post, or ...)? I personally think this would be great. I would still use Google for general-purpose search, but a lot of my search needs are trying to remember that really interesting article someone posted to HN a year ago that is germane to what I'm doing now.

(Author here)

One of the things that confused me is that regression models can be predictive, just like time series forecasting — they just do so in a different way. I tried to make this clear in the article (or maybe I’m not understanding what you’re saying).

In a regression model, you’re predicting target variables from feature variables. In a time series, you’re predicting the same variable from its past behavior. This is a subtle but crucial difference.

(And then you can do time series with covariates, which combines the two.)

(Author here). Or a lot of trial & error :).

My kids have school-issued devices, and there is a ton of tracking on them.

I set up my Pi-Hole on a Raspberry Pi Zero W. I have some brief notes here on my setup: https://www.thelis.org/blog/pi-hole. It works well.

Are you using slots (https://cloud.google.com/bigquery/docs/slots)? If you aren't, I'd highly recommend you switch. My guess is that it would make your costs much more predictable (it did for us).

Note that this is not the default! :-)