Prasad Chalasani
@pchalasani.bsky.social
Building:
Langroid - Multi-Agent LLM framework: https://github.com/langroid/langroid
Productivity tools for Claude-Code & other CLI agents:
https://github.com/pchalasani/claude-code-tools
IIT CS, CMU/PhD/ML.
Ex- ASU, Los Alamos, Goldman Sachs, Yahoo
Langroid - Multi-Agent LLM framework: https://github.com/langroid/langroid
Productivity tools for Claude-Code & other CLI agents:
https://github.com/pchalasani/claude-code-tools
IIT CS, CMU/PhD/ML.
Ex- ASU, Los Alamos, Goldman Sachs, Yahoo
Fun new game: prompt gymnastics to get past Sora’s content filter 😂 Here’s one result:
October 7, 2025 at 1:43 AM
Fun new game: prompt gymnastics to get past Sora’s content filter 😂 Here’s one result:
With the Claude Sonnet 4.5 release there are interesting new context management and memory features in the API —
www.anthropic.com/news/context...
Anyone know if these are used in Claude-Code?
www.anthropic.com/news/context...
Anyone know if these are used in Claude-Code?
Managing context on the Claude Developer Platform
We're introducing new capabilities for managing your agents’ context on the Claude Developer Platform: context editing and the memory tool.
www.anthropic.com
September 30, 2025 at 12:18 PM
With the Claude Sonnet 4.5 release there are interesting new context management and memory features in the API —
www.anthropic.com/news/context...
Anyone know if these are used in Claude-Code?
www.anthropic.com/news/context...
Anyone know if these are used in Claude-Code?
Used Claude Code to build a simple tool tmux-cli that lets CC puppet your terminal -- i.e. interact with CLIs expecting user input -- useful to let CC autonomously run debuggers, test CLI/UI scripts, spawn other Claudes (fully visible unlike black-box sub-agents) github.com/pchalasani/c...
July 31, 2025 at 5:53 PM
Used Claude Code to build a simple tool tmux-cli that lets CC puppet your terminal -- i.e. interact with CLIs expecting user input -- useful to let CC autonomously run debuggers, test CLI/UI scripts, spawn other Claudes (fully visible unlike black-box sub-agents) github.com/pchalasani/c...
Reposted by Prasad Chalasani
🤯 I was asked internally to summarize @thursdai_pod in 1 slide and just noticed, that just after 2 years of the live shows, 1.5 years of podcast releases and almost 100 episodes (this week is #99!) we have just crossed...
100,000 downloads on Sub**ack 🔥
100,000 downloads on Sub**ack 🔥
April 2, 2025 at 10:28 PM
🤯 I was asked internally to summarize @thursdai_pod in 1 slide and just noticed, that just after 2 years of the live shows, 1.5 years of podcast releases and almost 100 episodes (this week is #99!) we have just crossed...
100,000 downloads on Sub**ack 🔥
100,000 downloads on Sub**ack 🔥
February 2, 2025 at 12:56 PM
Thanks to a Kagi subscription I haven’t seen a sponsored search ad in a really long time — so long I can’t even remember what those look like 🤣 It’s 2025 and I’m shocked how many still put up with ad surveillance. It’s high time to neuter the business models of Mega-Corps founded on this.
"... the main reason most people are still using Google is they're just not aware of the alternatives."
How do we make more people aware of Kagi?
Kagi's commitment to an ad-free experience extends to our own marketing. We rely purely on word-of-mouth support.
Will you help us spread the word? 🙏
How do we make more people aware of Kagi?
Kagi's commitment to an ad-free experience extends to our own marketing. We rely purely on word-of-mouth support.
Will you help us spread the word? 🙏
My wife just complained that her Google image search results are completely overrun by gen AI.
I compared my #kagi results for the same query and it was a completely different story. And that was even before turning on the AI filter.
I really do think the main reason most people are still […]
I compared my #kagi results for the same query and it was a completely different story. And that was even before turning on the AI filter.
I really do think the main reason most people are still […]
January 13, 2025 at 12:54 AM
Thanks to a Kagi subscription I haven’t seen a sponsored search ad in a really long time — so long I can’t even remember what those look like 🤣 It’s 2025 and I’m shocked how many still put up with ad surveillance. It’s high time to neuter the business models of Mega-Corps founded on this.
uv is great for python packaging. One of several benefits: You can publish your lib to PyPi so people can run specific scripts from anywhere with zero installs/venv/repo-cloning using
uvx --from lib script
You just need to add a [project.scripts] section with paths
to the scripts
uvx --from lib script
You just need to add a [project.scripts] section with paths
to the scripts
GitHub - langroid/langroid-examples: Using Langroid's Multi-Agent Framework to Build LLM Apps
Using Langroid's Multi-Agent Framework to Build LLM Apps - langroid/langroid-examples
github.com
December 15, 2024 at 4:56 PM
uv is great for python packaging. One of several benefits: You can publish your lib to PyPi so people can run specific scripts from anywhere with zero installs/venv/repo-cloning using
uvx --from lib script
You just need to add a [project.scripts] section with paths
to the scripts
uvx --from lib script
You just need to add a [project.scripts] section with paths
to the scripts
QwQ-32B-Preview from Qwen is the best "o1-like" LLM today. You can run it remotely via glhf.chat super-fast endpoint.
Langroid lets you easily use it -- it gets the "Sally question" right, after a LOT of thinking 🤔 (images below)
Langroid tour: langroid.github.io/langroid/tut...
Langroid lets you easily use it -- it gets the "Sally question" right, after a LOT of thinking 🤔 (images below)
Langroid tour: langroid.github.io/langroid/tut...
December 7, 2024 at 1:38 PM
QwQ-32B-Preview from Qwen is the best "o1-like" LLM today. You can run it remotely via glhf.chat super-fast endpoint.
Langroid lets you easily use it -- it gets the "Sally question" right, after a LOT of thinking 🤔 (images below)
Langroid tour: langroid.github.io/langroid/tut...
Langroid lets you easily use it -- it gets the "Sally question" right, after a LOT of thinking 🤔 (images below)
Langroid tour: langroid.github.io/langroid/tut...
llama3.3-70b is a seriously impressive model. Langroid lets you easily try it via groq's insanely fast endpoint - just set the model name to "groq/llama-3.3-70b-specdec"
Langroid: agent-oriented LLM framework from CMU/UW-Madison researchers
Quick tour: langroid.github.io/langroid/tut...
Langroid: agent-oriented LLM framework from CMU/UW-Madison researchers
Quick tour: langroid.github.io/langroid/tut...
December 7, 2024 at 1:05 AM
llama3.3-70b is a seriously impressive model. Langroid lets you easily try it via groq's insanely fast endpoint - just set the model name to "groq/llama-3.3-70b-specdec"
Langroid: agent-oriented LLM framework from CMU/UW-Madison researchers
Quick tour: langroid.github.io/langroid/tut...
Langroid: agent-oriented LLM framework from CMU/UW-Madison researchers
Quick tour: langroid.github.io/langroid/tut...
Reposted by Prasad Chalasani
✍️ "Hard truths about AI-assisted coding" tips & tricks in my latest article: bit.ly/ai-assisted
While AI-Assisted coding can get you 70% of the way there (great for prototypes or MVPs), the final 30% requires significant human intervention for quality and maintainability.
While AI-Assisted coding can get you 70% of the way there (great for prototypes or MVPs), the final 30% requires significant human intervention for quality and maintainability.
December 5, 2024 at 12:07 AM
✍️ "Hard truths about AI-assisted coding" tips & tricks in my latest article: bit.ly/ai-assisted
While AI-Assisted coding can get you 70% of the way there (great for prototypes or MVPs), the final 30% requires significant human intervention for quality and maintainability.
While AI-Assisted coding can get you 70% of the way there (great for prototypes or MVPs), the final 30% requires significant human intervention for quality and maintainability.
Reposted by Prasad Chalasani
Tomorrow morning at the Simons Institute : Sparks versus embers, can LLMs solve major open mathematical conjectures?
(FWIW I agree with everything in the Embers paper, so I guess the debate will be about the conclusions to draw from current evidence!)
simons.berkeley.edu/talks/sebast...
(FWIW I agree with everything in the Embers paper, so I guess the debate will be about the conclusions to draw from current evidence!)
simons.berkeley.edu/talks/sebast...
Debate: Sparks versus embers
Debaters: Sebastien Bubeck (OpenAI), Tom McCoy (Yale) Discussants: Pavel Izmailov (Anthropic), Ankur Moitra (MIT) Moderator: Anil Ananthaswamy
simons.berkeley.edu
December 4, 2024 at 6:44 PM
Tomorrow morning at the Simons Institute : Sparks versus embers, can LLMs solve major open mathematical conjectures?
(FWIW I agree with everything in the Embers paper, so I guess the debate will be about the conclusions to draw from current evidence!)
simons.berkeley.edu/talks/sebast...
(FWIW I agree with everything in the Embers paper, so I guess the debate will be about the conclusions to draw from current evidence!)
simons.berkeley.edu/talks/sebast...
This is a fun and highly informative weekly show to watch/hear, to catch up on LLM-related developments. Run by @altryne.bsky.social with commentary by @wolfram.ravenwolf.ai @yampeleg.bsky.social @nisten.bsky.social
🔥 This week's episode was indeed special, we chatted about the first open source intelligence model with @JustinLin610 from the @Alibaba_Qwen team, covered the H runner from @hcompany_ai , Olmo from @allen_ai and Hymba 1.5B from @nvidia
Give this a watch/listen 👇
Give this a watch/listen 👇
December 3, 2024 at 7:45 PM
This is a fun and highly informative weekly show to watch/hear, to catch up on LLM-related developments. Run by @altryne.bsky.social with commentary by @wolfram.ravenwolf.ai @yampeleg.bsky.social @nisten.bsky.social
Reposted by Prasad Chalasani
A special LLM course focusing on Smol models, the <2B parameters generic models which can run on-device/in-browser
For anyone interested in fine-tuning or aligning LLMs, I’m running this free and open course called smol course. It’s not a big deal, it’s just smol.
🧵>>
🧵>>
December 3, 2024 at 11:51 AM
A special LLM course focusing on Smol models, the <2B parameters generic models which can run on-device/in-browser
Reposted by Prasad Chalasani
How do LLMs learn to reason from data? Are they ~retrieving the answers from parametric knowledge🦜? In our new preprint, we look at the pretraining data and find evidence against this:
Procedural knowledge in pretraining drives LLM reasoning ⚙️🔢
🧵⬇️
Procedural knowledge in pretraining drives LLM reasoning ⚙️🔢
🧵⬇️
November 20, 2024 at 4:35 PM
How do LLMs learn to reason from data? Are they ~retrieving the answers from parametric knowledge🦜? In our new preprint, we look at the pretraining data and find evidence against this:
Procedural knowledge in pretraining drives LLM reasoning ⚙️🔢
🧵⬇️
Procedural knowledge in pretraining drives LLM reasoning ⚙️🔢
🧵⬇️
Reposted by Prasad Chalasani
I put together all of the Bluesky tools/resources I've collected into a list on my website. There are over 60+ different links now, and I've organized them by category: dame.blog/skeet-tools
Enjoy! 🦋
Enjoy! 🦋
November 24, 2024 at 1:49 AM
I put together all of the Bluesky tools/resources I've collected into a list on my website. There are over 60+ different links now, and I've organized them by category: dame.blog/skeet-tools
Enjoy! 🦋
Enjoy! 🦋
Reposted by Prasad Chalasani
Reposted by Prasad Chalasani
Reverse Thinking Makes LLMs Stronger Reasoners
abs: arxiv.org/abs/2411.19865
Train an LLM to be able to generate forward reasoning from question, backward question, and backward reaoning from backward question
Shows an average 13.53% improvement over the student model’s zero-shot performance
abs: arxiv.org/abs/2411.19865
Train an LLM to be able to generate forward reasoning from question, backward question, and backward reaoning from backward question
Shows an average 13.53% improvement over the student model’s zero-shot performance
December 2, 2024 at 10:45 AM
Reverse Thinking Makes LLMs Stronger Reasoners
abs: arxiv.org/abs/2411.19865
Train an LLM to be able to generate forward reasoning from question, backward question, and backward reaoning from backward question
Shows an average 13.53% improvement over the student model’s zero-shot performance
abs: arxiv.org/abs/2411.19865
Train an LLM to be able to generate forward reasoning from question, backward question, and backward reaoning from backward question
Shows an average 13.53% improvement over the student model’s zero-shot performance
Reposted by Prasad Chalasani
Starter pack for open-source maintainers in the Python ecosystem! Feel free to share and let me know who I'm missing!
go.bsky.app/D2Be5mg
go.bsky.app/D2Be5mg
November 21, 2024 at 9:41 AM
Starter pack for open-source maintainers in the Python ecosystem! Feel free to share and let me know who I'm missing!
go.bsky.app/D2Be5mg
go.bsky.app/D2Be5mg
Reposted by Prasad Chalasani
When large language models surpass neuroscientists for prediction of results of experiments
nature.com/articles/s41...
"which presages a future where LLMs assist humans
in making discoveries. Our approach is not neuroscience specific and is transferable to other knowledge-intensive endeavours"
nature.com/articles/s41...
"which presages a future where LLMs assist humans
in making discoveries. Our approach is not neuroscience specific and is transferable to other knowledge-intensive endeavours"
November 27, 2024 at 3:09 PM
When large language models surpass neuroscientists for prediction of results of experiments
nature.com/articles/s41...
"which presages a future where LLMs assist humans
in making discoveries. Our approach is not neuroscience specific and is transferable to other knowledge-intensive endeavours"
nature.com/articles/s41...
"which presages a future where LLMs assist humans
in making discoveries. Our approach is not neuroscience specific and is transferable to other knowledge-intensive endeavours"
Reposted by Prasad Chalasani
The AT Protocol unlocks exciting possibilities:
- Building custom feeds using ML
- Creating dashboards for data exploration
- Developing custom models for Bluesky
To gather @bsky.app resources on @huggingface.bsky.social. I've established a community org 🤗 huggingface.co/bluesky-comm...
- Building custom feeds using ML
- Creating dashboards for data exploration
- Developing custom models for Bluesky
To gather @bsky.app resources on @huggingface.bsky.social. I've established a community org 🤗 huggingface.co/bluesky-comm...
bluesky-community (Bluesky Community)
Tools for Bluesky 🦋
huggingface.co
November 25, 2024 at 3:59 PM
The AT Protocol unlocks exciting possibilities:
- Building custom feeds using ML
- Creating dashboards for data exploration
- Developing custom models for Bluesky
To gather @bsky.app resources on @huggingface.bsky.social. I've established a community org 🤗 huggingface.co/bluesky-comm...
- Building custom feeds using ML
- Creating dashboards for data exploration
- Developing custom models for Bluesky
To gather @bsky.app resources on @huggingface.bsky.social. I've established a community org 🤗 huggingface.co/bluesky-comm...
one of the nice things about bsky is how easy it is to get data from it via the "atproto" python client. See the image and gist for example code, showing how you can get an author's posts, and get your own "home" timeline.
gist: gist.github.com/pchalasani/c...
gist: gist.github.com/pchalasani/c...
November 24, 2024 at 10:50 PM
one of the nice things about bsky is how easy it is to get data from it via the "atproto" python client. See the image and gist for example code, showing how you can get an author's posts, and get your own "home" timeline.
gist: gist.github.com/pchalasani/c...
gist: gist.github.com/pchalasani/c...
Reposted by Prasad Chalasani
Write math on 🦋 with UnicodeIt!
For example: θ ∈ ℝⁿ or pp̅ → μ⁺μ⁻
Use website or install system-wide in Linux, macOS, or windows
www.unicodeit.net
(Created several years ago with @svenkreiss.bsky.social)
For example: θ ∈ ℝⁿ or pp̅ → μ⁺μ⁻
Use website or install system-wide in Linux, macOS, or windows
www.unicodeit.net
(Created several years ago with @svenkreiss.bsky.social)
November 23, 2024 at 9:34 PM
Write math on 🦋 with UnicodeIt!
For example: θ ∈ ℝⁿ or pp̅ → μ⁺μ⁻
Use website or install system-wide in Linux, macOS, or windows
www.unicodeit.net
(Created several years ago with @svenkreiss.bsky.social)
For example: θ ∈ ℝⁿ or pp̅ → μ⁺μ⁻
Use website or install system-wide in Linux, macOS, or windows
www.unicodeit.net
(Created several years ago with @svenkreiss.bsky.social)