trq_

5 ms·

1.
by trq_ 25d ago
Hi all, Thariq from the Claude Code team here. I posted this on Twitter, but just reposting here: We sometimes test API serving configs in Claude Code before rolling them out, and one running now maps the numerical effort value differently.
2.
by trq_ 2mo ago
Thank you! Appreciate it. I love wrestling with the problem of making sure Claude is in the loop with you in the way you want. More to come!
3.
by trq_ 2mo ago
I did make the decision, and shipped the PR.
4.
by trq_ 2mo ago
automode mostly fixes these things, it runs a classifier on every request that would have required permissions to make sure it matches your request
5.
by trq_ 2mo ago
Hi everyone, It's Thariq from the Claude Code team here. This was my change! I made the AskUserQuestion tool so am generally in charge of maintaining it. First, overall wanted to apologize and agree that this did not meet our bar and d
6.
by trq_ 2mo ago
Hi, it's Thariq from the Claude Code Team here. Thanks for the detailed report. We’re confident this is a hallucination but of course take these reports seriously and the team is looking into it. We’ll report back if anything turns up.
7.
by trq_ 3mo ago
hi guys, this is my change- will patch it to be an opt-in
8.
by trq_ 4mo ago
Hi there, Thariq from the Claude team here. Sorry this is happening, we'll fix it ASAP. We don't want anyone to feel locked into the tool. Claude's designs are HTML/CSS/JS that any editor can handle; we'll make
9.
by trq_ 5mo ago
Hey everyone, Thariq from the Claude Code team. We've been on this since the bug surfaced. Everyone affected is getting a full refund and an extra grant of usage credits equal to their monthly subscription as our apology. You can see m
10.
by trq_ 5mo ago
Hi, it's Thariq from the Claude Code team here. Sorry to hear, was wondering if you could find a session where this happens and hit /feedback and just say something like stop hook not firing and we'll take a look.
11.
by trq_ 6mo ago
Claude Code is not an electron app.
12.
by trq_ 8mo ago
Yes, we do but harnesses are hard to eval, people use them across a huge variety of tasks and sometimes different behaviors tradeoff against each other. We have added some evals to catch this one in particular.
13.
by trq_ 8mo ago
Hi everyone, Thariq from the Claude Code team here. Thanks for reporting this. We fixed a Claude Code harness issue that was introduced on 1/26. This was rolled back on 1/28 as soon as we found it. Run `claude update` to make sure
14.
by trq_ 9mo ago
Hi, work on Claude Code here! Let me know if you have any feedback!
15.
by trq_ 10mo ago
We're back up! It was about ~30 minutes of downtime this morning, our apologies if it interrupted your work.
16.
Show HN: Write Stories by Steering a LLM(latentlit.goodfire.ai)
2 points by trq_ 2y ago|0 comments
17.
LLM-Powered Sorting with TrueSkill(thariq.io)
5 points by trq_ 2y ago|1 comments
18.
Show HN: Opensourcing Sparse Autoencoders for Llama 3.3 70B(huggingface.co)
1 points by trq_ 2y ago|0 comments
19.
by trq_ 2y ago
Hmm the hallucination would happen in the auto labelling, but we review and test our labels and they seem correct!
20.
by trq_ 2y ago
If you're hacking on this and have questions, please join us on Discord: https://discord.gg/vhT9Chrt
21.
by trq_ 2y ago
We haven't yet found generalizable "make this model smarter" features, but there is a tradeoff of putting instructions in system prompts, e.g. if you have a chatbot that sometimes generates code, you can give it very specific
22.
Show HN: Llama 3.3 70B Sparse Autoencoders with API access(goodfire.ai)
201 points by trq_ 2y ago|51 comments
23.
Should Developers care about AI Interpretability?(thariq.io)
10 points by trq_ 2y ago|0 comments
24.
by trq_ 2y ago
This is incredible! I haven't seen that repo yet, thank you for pointing it out, and the writing
25.
by trq_ 2y ago
Yeah, I think the idea of finding out what flavor of uncertainty you have is very interesting.
26.
by trq_ 2y ago
This is awesome, can't wait for evals against Claude Computer Use!
27.
by trq_ 2y ago
Yeah! I want to use the logprobs API, but you can't for example: - sample multiple logits and branch (we maybe could with the old text completion API, but this no longer exists) - add in a reasoning token on the fly - stop execution, a
28.
by trq_ 2y ago
I want to build intuition on this by building a logit visualizer for OpenAI outputs. But from what I've seen so far, you can often trace down a hallucination. Here's an example of someone doing that for 9.9 > 9.11: https:&#x2F
29.
by trq_ 2y ago
I mean, LLMs certainly know representations of what words means and their relationship to each other, that's what the Key and Query matrices hold for example. But in this case, it means that the underlying point in embedding space does
30.
by trq_ 2y ago
Yeah wouldn't be surprised if the big labs are doing more than just arg max in the sampling.

More ›