jbellis

12 ms·

151.
by jbellis 1y ago
that's fair! we thought that at 23m it was already pushing attention spans pretty hard :)
152.
by jbellis 1y ago
I would have said that Harper Reed's workflow (brainstorm spec, then co-plan a plan, then execute using LLM codegen) is basically best practice today and I'm surprised that the author adds that "I’ve not been successful using
153.
by jbellis 1y ago
My mistake, I was in fact looking at one of the linked details pages
154.
by jbellis 1y ago
not sure wym, it's in the headline of the article that Opus 4 has 200k context (same as sonnet 3.7 with the beta header)
155.
by jbellis 1y ago
Good, I was starting to get uncomfortable with how hard Gemini has been dominating lately ETA: I guess Anthropic still thinks they can command a premium, I hope they're right (because I would love to pay more for smarter models). >
156.
by jbellis 1y ago
Grok 3 mini is the best model in its price range for code, that doesn't train on your data. So it's part of Brokk's free plan. https://brokk.ai
157.
by jbellis 1y ago
iirc these benchmarks have such short runtimes that they are not useful for assessing real-world java jit performance
158.
by jbellis 1y ago
more like 100x faster than Python/Ruby and that's before you throw in real multithreading
159.
by jbellis 1y ago
On my list of programmers I want to meet before it's too late, along with Brian Kernighan.
160.
by jbellis 1y ago
Short answer: static analysis Long answer: https://brokk.ai/blog/lean-context-lightning-development
161.
by jbellis 1y ago
Confession time: litellm still doesn't support parallel tool calls with Gemini models [ https://github.com/BerriAI/litellm/issues/9686 ] so we wrote our own "parallel tool calls" on top of Struct
162.
by jbellis 1y ago
Yes! Han Xiao at Jina wrote a great article that goes into a lot more detail on how to turn this into a production quality agentic search: https://jina.ai/news/a-practical-guide-to-implementing-deeps... This is the sam
163.
by jbellis 1y ago
you need to know its architecture to run inference against it possibly this is buried in the Chromium source if it's part of the OSS personality, if not difficulty of reverse engineering would be pretty high
164.
by jbellis 1y ago
All the AR glasses vendors have a pocketable dongle for this.
165.
by jbellis 1y ago
TIL that Chrome ships an internal embedding model, interesting! It's a shame that it's not open source, unlikely that there's anything super proprietary in an embeddings model that's optimized to run on CPU. (I'd us
166.
by jbellis 1y ago
[abstract] This approach significantly reduces the KV cache size relative to traditional multi-head attention [3.3] For saving the KV cache, only the intermediate latent representations need to be stored: [latex] where r is much smaller tha
167.
by jbellis 1y ago
they're both useful search is an active "I'm looking for X" related articles is a passive "hey thanks for reading this article, you might also like Y"
168.
by jbellis 1y ago
Great to see embeddings getting some love outside the straight-up-ML space! I had a non-traditional use case recently, as well. I wanted to debounce the API calls I'm making to gemini flash as the user types his instructions, and I de
169.
by jbellis 1y ago
You're right, I did like it. Thanks!
170.
by jbellis 1y ago
usually OpenAI, but yes
171.
by jbellis 1y ago
That just means it's using Sonnet, not that it's using Claude Code.
172.
by jbellis 1y ago
I love Aider, but I got frustrated with its limitations and ended up creating Brokk to solve them: https://brokk.ai/ Compared to Aider, Brokk - Has a GUI (I know, tough sell for Aider users but it really does help when mana
173.
by jbellis 1y ago
Called out by one of my favorite symphonic metal albums, featuring a Swedish band with an American singer covering French pop songs in metal covers. https://en.m.wikipedia.org/wiki/Les_Fleurs_du_Mal_(Therion_a...
174.
by jbellis 1y ago
This is particularly impressive because polymarket failed harder than I can remember it ever doing at predicting the Pope https://www.reddit.com/r/slatestarcodex/s/PRqb1nBVhA
175.
by jbellis 1y ago
Wow, 2001. Legitimately prescient. And verification ("evaluation" we call it now) really is the key, although most people working on "AI apps" haven't figured it out yet. Follow Hamel to catch up on the state of the
176.
by jbellis 1y ago
what do you see as the part that you do better than everyone else in this space?
177.
by jbellis 1y ago
the core problem is that despite 50 years of effort, software development remains defiantly unquantifiable along almost every dimension until we make progress on that, things like building a technical debt portfolio will be doomed to failur
178.
by jbellis 1y ago
Is there a way to find out what the rate limits are before you hit them?
179.
by jbellis 1y ago
this is true as far as it goes, if you don't use it you do in fact lose it but i've convinced very very good engineers to switch from vim to a "real" idea and their productivity went through the roof do they have a worse
180.
by jbellis 1y ago
I've heard this idea from multiple smart people but spec to code with an LLM takes something like six orders of magnitude more work than a traditional compiler, solving two of those OOMs with faster GPUs just doesn't get you there

More ›