pcwelder

7 ms·

1.
by pcwelder 25d ago
The whole thing (article, benchmark) is a slop soup. - Metaphor overload. We get it, it's just like car racing. Show some mercy on human readers. - X, not Y - A, never B - Tasteless em dashes - Hallucinated data, like model add date. T
2.
by pcwelder 2mo ago
The most I enjoy working with AI is my special workflow. I ask it to plan the feature in a separate worktree. In parallel I start coding without being biased by AI and vice versa. At some point I read its plan and iterate on it all the whil
3.
by pcwelder 3mo ago
Could be json or non json. Instead of using tools in API, you ask model to share structured output in text. You parse the string to get the JSON. Gives much more control over things you can do. For example model shares <tool_call name=&q
4.
by pcwelder 3mo ago
https://news.ycombinator.com/item?id=48496895 Found the time traveller.
5.
MiniMax-M3: A native multimodal model with 1M context(huggingface.co)
10 points by pcwelder 3mo ago|0 comments
6.
by pcwelder 3mo ago
Great! Finally follows custom tool call format (k2.6 couldn't). It's a good indicator of instructions following and agentic behaviour. UIs it's generating is pretty good, not without problems, but certainly better than other
7.
by pcwelder 4mo ago
None of the deepseek models are multimodal. How are you guys able to use it in daily work without image input? For example it's just so natural to share screenshots in a chat.
8.
by pcwelder 4mo ago
Opus 4.7 selects such palette and motifs by default. Might even be first iteration of claude design.
9.
by pcwelder 4mo ago
Opus is not the correct tier to compare this flash model with. On my tasks it has not been as good as even Sonnet 4.6 so far. Instruction following over long context feels worse. It's not a bad model by any means, better than any pro o
10.
by pcwelder 4mo ago
It's worth noting that Claude Code itself doesn't use the `insert` tool. (It also uses custom edit tool not the suite's predefined str_replace) Also as a person developing agentic code tools since before Claude Code, I'm
11.
by pcwelder 7mo ago
To sonnet 4.6 if you tell it first that "You're being tested for intelligence." It answers correctly 100% of the times. My hypothesis is that some models err towards assuming human queries are real and consistent and not out
12.
by pcwelder 7mo ago
Great work, but concurrency is lost. With search-replace you could work on separate part of a file independently with the LLM. Not to mention with each edit all lines below are shifted so you now need to provide LLM with the whole content.
13.
by pcwelder 7mo ago
Cool! Please share your work if possible! I couldn't decide on folding and reducing noise so I'm stuck on that front. I believe there is some elegant solution that I'm missing, hope to see your take.
14.
by pcwelder 7mo ago
All anthropic models. Gemini 2.5 pro and above. Gemini 3 flash is very good too. GPT models can follow tool format correctly but don't keep on going. Grok-4+ are decent but with issues in longer chats. Kimi 2.5 has issues with it rever
15.
by pcwelder 7mo ago
I had added z-ai in allow list explicitly and verified that it's the one being used.
16.
by pcwelder 7mo ago
It's live on openrouter now. In my personal benchmark it's bad. So far the benchmark has been a really good indicator of instruction following and agentic behaviour in general. To those who are curious, the benchmark is just the a
17.
by pcwelder 7mo ago
Each repost is worth it. This, along with John Ousterhout's talk [1] on deep interfaces was transformational for me. And this is coming from a guy who codes in python, so lots of transferable learnings. [1] https://www.youtu
18.
by pcwelder 8mo ago
> These are sending all files it can access TBF, Cursor's code indexing works the same way, it has to send all workspace files to their servers. Auto-completion systems need previous edits to suggest next edits so no surprises their
19.
by pcwelder 8mo ago
Sonnet has the same behavior: drops thinking on user message. Curiously in the latest Opus they have removed this behavior and all thinking tokens are preserved.
20.
by pcwelder 9mo ago
``` from anthropic.types import MessageParam data: list[MessageParam] = [{"role": "user", "content": [{"type": "text", "text": ""}]}] ``` This for example works both in m
21.
by pcwelder 9mo ago
Displaying inferred types inline is a killer feature (inspired from rust lang server?). It was a pleasant surprise! It's fast too as promised. However, it doesn't work well with TypedDicts and that's a show-stopper for us. Ho
22.
by pcwelder 9mo ago
To those who are not deterred and feel yolo mode is worth the risk, there are two patterns that should perk your ears up. - Cleanup or deletion tasks. Be ready to hit ctrl c anytime. Led to disastrous nukes in two reddit threads. - Errors i
23.
by pcwelder 9mo ago
It just fetched the HTML and replicated it. The usage of table is a giveaway. Any LLM with browser tool can do it (Kombai one shots it too for example), because it's just cheating.
24.
by pcwelder 9mo ago
But that's cheating because it then has the source code containing the table and its styles. I can confirm that this is what it does. And if you ask it to not use tables, it cleverly uses div with the same layout as the table instead.
25.
by pcwelder 10mo ago
In RNNs and Transformers we obtain probability distribution of target variable directly and sample using methods like top-k or temprature sampling. I don't see the equivalence to MCMC. It's not like we have a complex probability f
26.
by pcwelder 11mo ago
I ϲаn guаrаntее thаt thе ОСR ϲаn't rеаd thіs sеntеnсе ϲоrrесtlу.
27.
by pcwelder 11mo ago
There are many unicode characters that look alike. There are also those zero width characters.
28.
by pcwelder 1y ago
It doesn't throw error in the REPL though. Surely you meant to share some other example?
29.
by pcwelder 1y ago
>if they don't whatever happens, happens What happens is you get an error. So you immediately know something is wrong. Javascript goes the extra mile to avoid throwing errors. So you've 3>"2" succeeding in Javascri
30.
by pcwelder 1y ago
Navigate to ` https://claude.ai/settings/data-privacy-controls ` and disable it before Sept 28. Isn't applicable to team plan.

More ›