Talha Chowdhury
128 posts
Talha Chowdhury
@mdtalhachy
building @ flowsurf | ex @ amazon, autodesk
Katılım Mart 2017
619 Takip Edilen957 Takipçiler
I’m creating a group of the most cracked devrels in SF
if you are:
- dev rel
- dev tool's marketing/ecosystem
- hosts the best hackathon/developer meetup
comment below! i will add you to the group : )
English
Running a full scale research team to build world's best Physics Simulation Engine for a particular project.
Using @claudeai's new Agent Teams configs.
Spent quite a time building my own agent team, figuring configuration for each team member.
This would be years of work in another pre-CC universe.
Life is good.
English
@sawyerhood quietly became beast of an IDE, can't recommend enough
English
@metapreston Figma make has pretty good UI generation scaffolding.
English
@ylecun @SebastienBubeck this gonna be all over my fb feed today 😇
English
@SebastienBubeck Best research environment?
Except for the whole "don't tell anyone about your research" part.
Research in secret is not research.
English
I've been in lots of places in my career. OAI is simply the best research environment I have ever seen. It's a combination of the field itself being a research gold mine + having access to the right mining tools + (most importantly) the freedom to explore. It's special.
Mark Chen@markchen90
How does OpenAI balance long-term research bets with product-forward research fundamentals? I’ve been getting this question a lot lately, usually framed as a suggestion that Jakub (@merettm) and I are pushing an increasingly product-focused agenda. That characterization is simply wrong. Foundational research has been core to OpenAI from the start, and today we run a research program with hundreds of exploratory projects - much like the ones that led to our reasoning-model breakthrough. The majority of our compute is allocated to foundational research and exploration - and not product milestones. Anyone who has spent time with me or Jakub knows we are the last people in the world who would push for the advancement of products over the advancement of research. We’re in the business of creating an automated scientist, and capabilities that were considered grand challenges just a few years ago (like IMO-level mathematical reasoning) now emerge as normal parts of the research process. We’re also seeing our models accelerate researchers worldwide, helping advance work across biology, mathematics, physics, and even our own research. Jakub and I put a lot of effort into ensuring that research stays focused on uncovering algorithms that will scale to the compute we’ll have a year from now. We protect mindshare and amplify discourse on exploratory work. We do this while recognizing that we’re also a deployment company - and that deployment gives us access to even larger-scale compute, richer feedback, and more room for exploration. Our researchers are passionate about having their work out in the world, and a special slice of our org is dedicated to making sure our deployments are delightful for end users. Our goal isn’t to turn research into a quarterly race. It’s to build a durable research engine - one that compounds learning over time and consistently turns long-horizon exploration into real, measurable advances, while ensuring those advances become valuable in the real world. That’s the roadmap we’re executing on. And while there have been ups and downs over the last decade (as you expect with any research program), I think most of our researchers would share my strong optimism today.
English
@n0w00j built pinpoint.sh lets you create specs for undescribeable complex bugs & features
English
I've been shipping at 100x speed since I've been using this workflow.
This will literally supercharge you if you're a builder, designer or anyone who's frustrated with AI not getting the design/specs that you have in your mind.
Launch this week. Meet Pinpoint: Just Pinpoint it.
English
Excited to announce AI Research Skills - an open-source library of 82 specialized skills for AI coding tools.
One command gives your agent expert knowledge in:
→ Model training & fine-tuning (TRL, Unsloth ...)
→ Distributed systems (DeepSpeed, FSDP ...)
→ Inference optimization (vLLM, TensorRT ...)
→ Agent building ? (Langchain, AutoGPT ...)
Works with Claude Code, Cursor, Gemini CLI, Windsurf, and Codex with one click interactive installation
@ npx @orchestra-research/ai-research-skills
If you found the ML paper writing skill useful, check out the comprehensive collection
github.com/Orchestra-Rese…
English
I'm an hiring an ops generalist to work closely with me in scaling Taxwire this year.
DMs open if this is you
English
for anyone who didn't make it to new media, hmu in DMs with your submission.
i'd love to chat with you
Brent Liang@liangsays
the 1st cohort of a16z new media fellows is here thank you to everyone who applied. the cultural power this group holds is immense grateful and looking forward to the 8 weeks ahead a16z.news/p/meet-the-a16…
English
Read the full findings in the blog post: talhachowdhury.com/posts/question…
English
Here's two surprising findings so far from my research from building QuestionBench:
1) An Oracle agent with direct access to all user information only achieved 68.40% success, proving that having information isn't enough without reasoning about relevance.
(2) GPT-5.2 with memory achieved 87.13% question appropriateness (vs Claude's 55-60%) while asking only 2.1 questions instead of 5, suggesting fundamentally different strategic approaches between models.
QuestionBench is a benchmark that tests frontier AI's ability to ask productive questions.
The OG problem I was investigating how good frontier AI models and tools are at asking questions that leads to task success?
It's worth looking into it because that's the path to truly autonomous AI agents or assistants.
English
If you're interested in RLVR, AI Evals, Memory, Egocentric Multimodal AI, or want to sponsor research in this area, get in touch!
English
The way a kid learns is by asking you incessant questions until they're satisfied. Overtime the number of questions reduce as they know more, and at one point they're capable of performing handed out tasks with maximum success because they have that knowledge graph they built. This is how true personal assistants will be built, too.
English
I'm deeply interested in solving the AGI problem from a multi disciplinary approach with a focus on building a truly helpful personal assistant that augments personal productivity. And I believe this is a big baby step towards that.
English
There are many other interesting insights that are coming up while investigating this and I'll keep sharing the updates.
English
Thought about it many times, lot of architectural barriers. For example new posts, resources on X that comes up post idea will need to be searched & indexed accordingly with each new post/resource triggering the indexing. Lots of computation. Two solution: 1) Need a company to do this at scale so a single index can be used for all users 2) Manually curate sources (such as HN) where scrapping new posts are ezz unlike X. Thoughts?
English
Too many ideas. Not enough time.
I want an app I can just send ideas to and have it collect resources/inputs over time (because I'm constantly being inspired). That way, when I do have a few spare moments, I have a nicely organized collection of "ideas and context" to start from.
It could also start planning some of them, and evolve the plan as I send it more things.
... And now I have another idea of something I'd love to build if I had more time 😅
English
The problem: new tools that are not well documented or recently open sourced projects on github aren't usually used by agentic coding tools when you ask it to choose best apis/implementation for the job. If you aren't aware of the tool and its tradeoffs and benefits, you might be missing out on many apis and implementations that might be better for your project. Some tool surfacing system that can integrate well with agentic coding tools would solve this pain.
English