Amir Yazdavar, Ph.D.

Founder of DylianAi, Ex-Lead AI Engineer @Optum, @Change_HC, @Bosch_AI @WeillCornell, @USC_ISI @NLM_LHC. DS,NLP,Gen AI, KG,Healthcare

[Andrej Karpathy @karpathy]

We'll be spending a lot more time trying to understand the outputs of language models. A few thoughts, tips & tricks:

Writing. Something I've had success with: Ask your LLM to explain something in ASD-STE100, it's a controlled language specification originally developed for aerospace maintenance documentation. LLMs well-versed in this language and it comes with heavy constraints on clean writing style that I often find a lot more readable. Sometimes I've tried to soften it a bit e.g. ask for "80% of the way to ASD-STE100" because the spec is quite stringent. But even better:

Diagrams / images. Instead of writing, ask your LLM to create a diagram. These can be a lot easier to process, parse, and understand. But even better:

Web pages. Ask for output "in HTML" to get a beautiful, interactive webpage. LLMs are getting really good at frontend and can create beautiful experiences, animations, etc. But even better:

Explainer videos. The output format I am most bullish on is fully custom / bespoke explainer videos generated on any arbitrary topic. Experiment with things like "Create a 3b1b style video explainer on X. Use my ElevenLabs API key for audio narration". (you'd need an API key for the latter or you can ask your LLM to find you decent free alternatives that use your local compute). This is actually starting to work!

In summary:
- As LLMs get better, they will do more and more of the legwork autonomously, and a lot more of our work will rise up the abstractions into oversight and understanding.
- Luckily, LLMs can help here too because as intelligence and code are increasingly abundant, you can ask for large, custom, discardable software artifacts (e.g. web apps, video explainers) that would have never made sense to create before. Push the boundaries here and you'll be surprised.

[Amelia hazel @ameliahazelai]

30 Websites That Feel "Illegal" But Are Perfectly Legal

1. https://t.co/eJvbiMP99g — Free unlimited AI image generation, quality rivals Midjourney

2. https://t.co/JTiEEEc9rW — Real-time AI image generation, draws as you go

3. https://t.co/t24R8CP81l — AI unlimited image upscaling, details auto-filled

4. https://t.co/AyV3zDUq5Z — AI one-click background removal/lighting fix/erasure

5. https://t.co/cB61YWYgjk — AI voice cloning, mimics any voice in 5 seconds

6. https://t.co/hewnbVy4np — Input lyrics to auto-generate full songs

7. https://t.co/KK68s5zgz5 — AI video generation pioneer, free Gen-3 trial

8. https://t.co/RPeUengsF7 — Kuaishou Keling AI, smoothest Chinese video generation

9. https://t.co/oruXJIB9pV — One photo + one audio clip = talking digital human

10. https://t.co/gfie9CzujB — Turns static photos into talking videos

11. https://t.co/dAh6O517jL — AI code writing, free quota enough for daily use

12. https://t.co/BTnSmDjE0N — Build websites by talking, zero code

13. https://t.co/85vct8qFZb — Vercel AI frontend generator, description becomes page

14. https://t.co/eEbu9s3EdR — Code in browser + AI assistance + one-click deploy

15. https://t.co/JCwk0inkQV — Paste text to auto-generate infographics

16. https://t.co/eJFFIraPnu — AI one-click PPT generation, say goodbye to PPT hell

17. https://t.co/oLMPtAZlYi — Notion AI for writing, summarizing, translating

18. https://t.co/Z3ehJImnOp — Sketch a drawing, AI generates real webpage

19. https://t.co/hZpjQSv8KG — AI search engine, answers any question instantly + sources

20. https://t.co/DzX9x85WSL — Developer-exclusive AI search, code issues solved on search

21. https://t.co/eN6D0Gd6uY — Real-time meeting transcription, free 300 minutes monthly

22. https://t.co/kNxOrDnT1r — AI short video editing, auto-finds highlight clips

23. https://t.co/JUsFhf8aQl — One site to use GPT-4o / Claude / Gemini

24. https://t.co/MPMHviKhqx — Open-source AI model free playground

25. https://t.co/I9R1C7EDCz — One-click AI background removal, 1-second output

26. https://t.co/GPkIB2zb3S — AI erases any object from photos

27. https://t.co/met9Y9bMZV — AI face swap, 1-minute turnaround

28. https://t.co/2bdwMQBNcR — AI music tagging + recommendations

29. https://t.co/x4rcZZcCvR — AI writing assistant, free version enough for daily use

30. https://t.co/n1nLjQe0b6 — Anthropic free AI assistant, top-tier long-text handling

Follow @ameliahazelai for more information

[Machina @EXM7777]

GPT-6 Astra inside Obsidian is honestly insane...

i built a knowledge base inspired by the concept of LLM Wiki from Andrej Karpathy, it stores everything my agents need for making long-form ads:

- winning ad structures mined from the Meta Ad Library
- the sales spine, extracted before a word gets written
- the script and the master soundtrack, mapped and cut
- character sheets locking characters, products and geography
- timed sequences ready for video generation
- the final montage and the reusable agent that runs it all

the full setup is in the article

[Andrej Karpathy @karpathy]

Software horror: litellm PyPI supply chain attack.

Simple `pip install litellm` was enough to exfiltrate SSH keys, AWS/GCP/Azure creds, Kubernetes configs, git credentials, env vars (all your API keys), shell history, crypto wallets, SSL private keys, CI/CD secrets, database passwords.

LiteLLM itself has 97 million downloads per month which is already terrible, but much worse, the contagion spreads to any project that depends on litellm. For example, if you did `pip install dspy` (which depended on litellm>=1.64.0), you'd also be pwnd. Same for any other large project that depended on litellm.

Afaict the poisoned version was up for only less than ~1 hour. The attack had a bug which led to its discovery - Callum McMahon was using an MCP plugin inside Cursor that pulled in litellm as a transitive dependency. When litellm 1.82.8 installed, their machine ran out of RAM and crashed. So if the attacker didn't vibe code this attack it could have been undetected for many days or weeks.

Supply chain attacks like this are basically the scariest thing imaginable in modern software. Every time you install any depedency you could be pulling in a poisoned package anywhere deep inside its entire depedency tree. This is especially risky with large projects that might have lots and lots of dependencies. The credentials that do get stolen in each attack can then be used to take over more accounts and compromise more packages.

Classical software engineering would have you believe that dependencies are good (we're building pyramids from bricks), but imo this has to be re-evaluated, and it's why I've been so growingly averse to them, preferring to use LLMs to "yoink" functionality when it's simple enough and possible.

[Morad Vaisi @RezaVaisi]

@elonmusk
Dear Elon,
In these critical days, the people of Iran need internet access. Please help them stay connected as they struggle to reclaim their country from the rule of the mullahs. Stand with the Iranian people. Iran will not forget its friends.

[Hossein Jahandideh @HosseinJahandi3]

@POTUS President Trump, I know you're not seriously going to negotiate with this terrorist regime and let them deceive you and continue terrorizing Iranians and the world. But just in case you need additional motivation to topple this regime, remember that this regime is throwing death threats at you every day. Take it personal.

[Harley Finkelstein @harleyf]

Like many of you I have been watching what is happening in Iran. The courage of the people there is unmistakable. They are asking for dignity. They are asking to live freely and safely. I admire their bravery and I stand with them.

In moments like this there are no complex sides. There is freedom or there is terror. The choice could not be more obvious.

[Sebastian Raschka @rasbt]

This interesting week started with DeepSeek V3.2!

I just wrote up a technical tour of the predecessors and components that led up to this:

🔗 https://t.co/JSAd9cx2s6

- Multi-Head Latent Attention
- RLVR
- Sparse Attention
- Self-Verification
- GRPO Updates

[Bojan Tunguz @tunguz]

Never had it better

[Greg Brockman @gdb]

an underrated skill in ML is deriving great insight from small wiggles in the graphs

[Sam Altman @sama]

gpt-oss is out!

we made an open model that performs at the level of o4-mini and runs on a high-end laptop (WTF!!)

(and a smaller one that runs on a phone).

super proud of the team; big triumph of technology.

[clem 🤗 @ClementDelangue]

Reflection-Llama-3.1-70B by @mattshumer_ @csahil28 @GlaiveAI is #1 trending on HF.

I've said it and will say it again: you don't need to be a big tech to fine-tune, optimize and run your own models for your specific constraints and you will benefit massively from it.

Ultimately, all tech organizations will do it, just the same way all of them write their own code!

[LangChain @LangChain]

Love @EdenEmarco177's take here:

"2024 is truly the year of the agents🤖

...

LangGraph🦜 is a cutting-edge flow engineering implementation that simplifies the creation and management of complex agentic flows."

[Maartje ter Hoeve @maartjeterhoeve]

I’m looking for a Ph.D. research intern in Human-Centered NLP to come work at Apple MLR (Machine Learning Research team), in Cupertino, CA! ☀️

Application details in thread below!

🧵👇

[1/2]

[Sarath Chandar @apsarathchandar]

I am recruiting MSc and PhD students for Fall 2024. Application deadline is December 01. Please apply through the @Mila_Quebec supervision request process here: https://t.co/gZAfne1g96. More details about the recruitment process here: https://t.co/DjjQj2Y2b6

[Yin Cui @YinCuiCV]

Our team is hiring research scientists and interns to advance generative AI and democratize content creation!

[Arsalan Mousavian @a__mousavian]

We have research internship positions in our team for the Summer of 2024. If you are passionate (and experienced) in using learning, LLM/VLMs, and 3D reasoning to take the robot manipulation capabilities to the next level, apply to our team: https://t.co/DcuK2x2Nts

[Nikolaos Sarafianos @sarafianosn]

The CoreAI team at Meta Reality Labs is hiring 2024 interns to work on 3D Generative AI. Joins us to explore (& publish) new methods for texture and shape synthesis and stylization. Feel free to message me if you have any questions or apply directly!

https://t.co/aQ6P8N6P0E

[Ankit Goyal @imankitgoyal]

Applications are open for PhD Research Intern in Robotics NVIDIA for Summer 2024. We are based in Seattle+Bay Area, led by Prof. Dieter Fox. Please consider applying!

Appication: https://t.co/bnwpnBtCXO

More info about our team can be found here:
https://t.co/YJi7XWyEvH

[Tanya Jonker @trjonker]

Internships for 2024! Come join us to spend some time sitting on the cutting edge of AR/VR and AI. We're a fun, dynamic, creative team, and internships have the opportunity to publish their original work. dm and/or apply directly.
https://t.co/laitslgdHs