Ankur Gupta
@getpy
Tweets on Python, Technology, Software Development, Programming. Writing book on DSPy - http://dspybook.com - Build AI Applications with Python and DSPy.
قد يعجبك
Optical illusion ... asked by my son to generate it. Signature and arguments for this image copy pasted as reference. #DSPy #NanoBananaPro
With microservices you need to pay the operational, cognitive and debugging complexity of the underlying network at every step. You need to implement resilience and fault tolerance - it's easy to say implement circuit breaker etc. but it's easier said than done. Parnas talked…
DSPy.rb 0.32.0 is out 🎉 Can't wait to get my hands on this! @kieranklaassen built a RubyLLM adapter that bridges @paolino's incredible work to DSPy.rb — unlocking 12+ LLM providers (Bedrock, VertexAI, DeepSeek,...) all through DSPy.rb's type-safe signatures. ```ruby gem…
Thank you @arnav_thebigman @kristahopsalong @michaelryan207 @ShangyinT @slimshetty_ @CShorten30 @ecardenas300 @thomasahle @cyrusofeden @krypticmouse @isaacbmiller1 @SVardhamanan @kcaverly_dev @paridhi2708 @rickbattlephoto @ammirsm — and @ChrisGPotts @matei_zaharia @heathercmiller
After a meteoric rise, DSPy is now the @stanfordnlp repository with the most GitHub stars. Big congratulations to @lateinteraction and his “team”. DSPy: Programming—not prompting—Foundation Models github.com/stanfordnlp/ds…
Back from Mumbai aka Bombay. It's still - City of dreams, hustle and humidity - It's organised chaos and something bengaluru can import - Skyline is 😍 and infrastructure has made a leap - People value time and rarely go back on their word unless external forces beyond their…
Every AI book becomes obsolete the moment it's published. And I should know, I've written one! Publishing contracts freeze your table of contents 6-12 months in advance. In AI, that's an eternity. Just 6 months ago, top models juggled 3-5 tool calls from a pool of 3-5 tools…
In today's episode of programming horror... In the Python docs of random.seed() def, we're told "If a is an int, it is used directly." [1] But if you seed with 3 or -3, you actually get the exact same rng object, producing the same streams. (TIL). In nanochat I was using the…
![karpathy's tweet image. In today's episode of programming horror...
In the Python docs of random.seed() def, we're told
"If a is an int, it is used directly." [1]
But if you seed with 3 or -3, you actually get the exact same rng object, producing the same streams. (TIL). In nanochat I was using the…](https://pbs.twimg.com/media/G7sc167aMAAuZi5.png)
If you struggle at validating your agent before deployment, OfficeQA is a great benchmark for this purpose! Somehow the dataset link is not very visible in the blog post... it's under the conclusion section, and here it is: github.com/databricks/off…
LLMs are claimed to reach PhD intelligence, but still fail mundane tasks. To understand this challenge, Databricks launched OfficeQA, a benchmark of useful tasks that require reliability&diligence, not specialized knowledge. We're also doing a competition! databricks.com/blog/introduci…
You log into an application and a chatbot interface is present as an option. How many chances do you give the chatbot to help you before you abandon it entirely?
We're working with the team to introduce it to pydantic AI.
Reliability is the barrier holding back agents, and right now the most common way to achieve it is scaling back ambitions. dbreunig.com/2025/12/06/the…
Folks managed to use GEPA right after it came out and then submit to a NeurIPS workshop, demonstrating 12.5% -> 62.5% gains on their task 📸
Terrific article on AI writing, its telltale signatures, the origins of its distinctiveness vs. human writing, and how humans are evolving our own writing in response. N.B.: I use a lot of em dashes in my casual writing, but am not an AI—I think. nytimes.com/2025/12/03/mag…
OpenEvolve's Widespread Adoption! Just 6 months after launch, OpenEvolve is being cited & used across academia & industry. Here's a breakdown of the amazing research built on top of it 👇
Someday Tejas Network can do this with the chips coming from Tata Electronics Fabs?.
China is finally starting to replace AMD/Intel processors in its communication systems (5G base stations) Phytium is the sole supplier for China Mobile's extra 5G base station procurement The chip is decent
TOON + DSPy = fewer tokens. Ran few benchmarks. TOON wins most tests. BAML only took one because of lower input tokens. Next up: quality-impact benchmarks + GEPA prompt-optimization experiments. And check my implementation of TOON adapter for DSPy.
United States الاتجاهات
- 1. Lakers 43.9K posts
- 2. Spurs 36.2K posts
- 3. Doug Dimmadome 8,748 posts
- 4. Godzilla 19.8K posts
- 5. Michigan 122K posts
- 6. Wemby 5,329 posts
- 7. Marcus Smart 2,138 posts
- 8. Sherrone Moore 62.8K posts
- 9. #Survivor49 5,098 posts
- 10. Dolly Parton 2,212 posts
- 11. #AEWDynamite 26.6K posts
- 12. Erika 175K posts
- 13. #PorVida 1,490 posts
- 14. Gabe Vincent 1,150 posts
- 15. Stephon Castle 3,612 posts
- 16. Jim Ward 9,145 posts
- 17. PETA 23.7K posts
- 18. Candace 218K posts
- 19. Gainax 3,552 posts
- 20. Vando 1,315 posts
قد يعجبك
-
PyCoder’s Weekly
@pycoders -
Scientific Python
@SciPyTip -
Python Software Foundation
@ThePSF -
Talk Python Podcast
@TalkPython -
Guido van Rossum
@gvanrossum -
Django
@djangoproject -
Real Python
@realpython -
Python Hub
@PythonHub -
Matt Makai | Full Stack Python | Plushcap
@fullstackpython -
PyCon US
@pycon -
PythonAnywhere
@pythonanywhere -
PyData
@PyData -
PyCharm, a JetBrains IDE
@pycharm -
Python Trending 🇺🇦
@pythontrending -
Michael Kennedy
@mkennedy
Something went wrong.
Something went wrong.