Antaripa Saha

tried going as much in detail as possible for the Colpali blog, it will be a long read antaripasaha.notion.site/ColPali-Effici…

@wh0sumit i sometimes wonder if i am missing out on something given i have been doing wfh from my hometown for the past 5 years of my career. then i self-reflect and say whatever and move on

i’ve been living in blr for over 1.5 years now, and i’ve realized i’ve only been to a handful of events. personally, i’ve always felt the information exchange at most events is pretty low. on the other hand, the people i’ve met 1:1 have been far more valuable for me. those conversations have taught me a lot. just my personal opinion.

@HamelHusain most of the medium articles and even many x articles nowadays

AI slop is not the same as AI generated You can make good stuff with AI. It’s just the low effort/quality stuff that’s slop. Writing has an endemic rn You can write well with AI if you take lots of care and edit heavily

@A_K_Nain no amount of change can fix the algo

If the algo is truly fixed, I expect an up tick in my reach as my mutuals are great!

New Blog Post: Do Automated Evals Work? There has been a rise of tools that look through your traces with AI and identifies issues. We tested these tools with real production data to see how good they are. Where they shine - They often spot issues human miss - Integrate into your workflow: viewing traces, creating LLM judges etc. Where they fall short - They miss problems that require domain expertise and taste - Don't have great mechanisms to learn from human feedback - You can get similar results from using your coding agent So you should use them? Yes, BUT do so iteratively with you in the loop. We describe how in the post: parlance-labs.com/blog/posts/aut… It's also a good idea to try using your coding agent with you in the loop, which we discuss in the post. This was written with @doesdatmaksense , who led the research and collated the results.

@neural_avb congratulations. one of the best high quality channel on ml out there. keep going!

Channel reached 33,000 subs today pizza is on me

@anmolbuildz @HamelHusain assuming you are talking about programmtic way of checking, this can be done if you have fixed heuristics to check on, what about cases where there's reasoning involved?

@doesdatmaksense @HamelHusain we bypass platforms and write pytest assertions on raw traces

Do automated evals actually work? I and @HamelHusain spent last few weeks testing the auto-evals efficacy of different evals platforms. We took 100 real production traces from an apartment-leasing voice agent, manually reviewed the failures, masked the labels, and asked different systems to do the same error-analysis task and discover failure modes. We tested dedicated eval platforms like Braintrust, Arize, and LangSmith, along with ChatGPT, Codex, Claude Code, and Factory Droid. Full post here: parlance-labs.com/blog/posts/aut…

@ThePeshwa @gridinoc @HamelHusain `By failures do you mean tool call failures or final outputs being wrong` -> both

@doesdatmaksense @gridinoc @HamelHusain By failures do you mean tool call failures or final outputs being wrong? If your labels were on final output correctness, did you have access to a sme for the final labels? Very relevant to my usecase and I struggle with access to SMEs.

@barrowjoseph yessir!

Writing a *long* post about speculative decoding, how it works and the major techniques developed through to dSpark. If you want to be a beta reader please let me know! 🙏 This goes for anyone who does or does not understand it yet!

@xeophon @HamelHusain 🫡

@doesdatmaksense @HamelHusain Great post!!

@_Suresh2 @HamelHusain do you mean the tool outputs too? we have provided the full traces bec many times the end outputs were correct but internal tool calls were wrong, so we wanted to check those failures too

@doesdatmaksense @HamelHusain did you also mask the intermediate reasoning? those steps often leak the failure mode.

@amuldotexe sir ap toh abhi chahe toh abhi hee banale

@doesdatmaksense Some day some day

presenting the home office built with adult money

@AnkitaxPriya @librahq Congratulations Ankita! Looking super cool in the hat

life update: joined @librahq :) gonna be building some cool AI tech, watch out this space 🧡

@pratty_agi thanks sir

@doesdatmaksense Goals for real. Congrats on the new milestone!

@deepakkttwt hey it’s made by my interior guy

@doesdatmaksense Link to the wall mounted shelf in second pic? Looks very clean btw!

@A_K_Nain thank you bhaiya. aayiye ap kabhi north east humare ghar

@doesdatmaksense Nice. Congrats! 💥💥

@jayitabhattac11 haha sure mam, anytime!

@doesdatmaksense oh damn I'll be shameless and ask for a house tour invite

@antoine_chaffin back to back banger releases (yesterday Tony, today you)!

Whether you are GPU poor or GPU rich, today's release of PyLate has something for you! GPU maxxers: MaxSim kernels greatly speed up training while lowering the memory requirements CPU enjoyers: TACHIOM enables lightning fast multi-vector indexing and search directly on CPU