301
Programming AIs worry me
For some inane reason, Github classifies me as a "major open source maintainer", which means I get a free copilot subscription.1 I've been using it for a...
302
Lessons from YC AI Startups
The evolution of the AI landscape, and YC's Copilot era.
303
ChatGPT has no inner monologue or meta-cognition
Update 2025: The post below describes ChatGPT and similar LLMs as they functioned at the time of writing in 2023. More recent systems do have a behind-the-scenes “inner monologue” which enables them t
304
A framework to securely use LLMs in companies – Part 1: Overview of Risks
Part 1 of a multi-part series on using LLMs securely within your organisation. This post provides a framework to categorize risks based on different use cases and deployment type.
305
Large Transformer Model Inference Optimization
[Updated on 2023-01-24: add a small section on Distillation.] Large transformer models are mainstream nowadays, creating SoTA results for a variety of tasks. They are powerful but very expensive to train and use. The extremely high inference cost, in both time and memory, is a big bottleneck for adopting a powerful…
306
The Decade of Deep Learning
As the 2010’s draw to a close, it’s worth taking a look back at the monumental progress that has been made in Deep Learning in this decade.[1] Driven by the development of ever-more powerful comput
307
Compiling ML models to C for fun
We make micrograd fly with a little compiler magic. In this post, we’ll write a ML compiler from scratch.
308
Sequential modeling enables scalable learning for large vision models
We introduce a novel sequential modeling approach which enables learning a Large Vision Model ( LVM ) without making use of any linguistic data. To do this, we define a common format, "visual sentences", in which we can represent raw images and videos as well as annotated data sources such as semantic segmentations…
309
The convolution empire strikes back
ConvNets Match Vision Transformers at Scale
310
I played chess against ChatGPT-4 and lost
GPT-4 will change the world.
311
312
The continued relevance of Weizenbaum’s warnings
“The computer has long been a solution looking for problems—the ultimate technological fix which insulates us from having to look at problems.” – Joseph Weizenbaum (1983) Trying to kee…
313
4000x Speedup in Reinforcement Learning with Jax
Web publication for "PureJaxRL"
314
How to Optimize a CUDA Matmul Kernel for CuBLAS-Like Performance: A Worklog
In this post, I’ll iteratively optimize an implementation of matrix multiplication written in CUDA.My goal is not to build a cuBLAS replacement, but to deepl...
315
We aren't close to creating a rapidly self-improving AI
The current paradigm will not lead to an overnight superintelligence explosion.
316
317
318
Multifaceted: The linguistic echo chambers of LLMs
This is a fun one. I’ve spent more time than I’d care to admit staring at LLM output. And there’s something that I’ve noticed: LLM-generated prose has a kind of… vibe. It’s difficult to describe,…
319
The AI hype bubble is the new crypto hype bubble
Back in 2017 Long Island Ice Tea – known for its undistinguished, barely drinkable sugar-water – changed its name to "Long Blockchain Corp." Its shares surged to a peak of 400% over their pre-announcement price. The company announced no specific integrations with any kind of blockchain, nor has it made any such…
320
AI model weight providers should not police uses, no matter how awful they are
The Electronic Frontier Foundation, a digital civil liberties advocacy group, recently published a good article entitled “ISPs Should Not Police Online Speech—No Matter How Awful It Is.” 1 It argues that providers within the internet “stack” should refrain from censoring traffic according to their own values. That…
321
Gptcommit: Never write a commit message again (with the help of GPT-3)
With gptcommit, you no longer have to spend time writing your commit messages. Install gptcommit and let GPT-3 do the writing for you.
322
323
How to Think Computationally about AI, the Universe and Everything
In his TED Talk, Stephen Wolfram covers the emergence of space by the application of computational rules to spacetime, gravity and quantum mechanics to AI and LLMs. Computational irreducibility and the ruliad.
324
What Will Transformers Transform?
Generative Pre-trained Transformer models (GPTs) are now all the rage and have inspired op-eds being written by everyone from Henry Kissinger (WSJ) to Noam Chomsky (NYTimes) in just the last month. That sure is some hype level.
325
The many ways that digital minds can know – A better way to think about LLMs
LLMs do all the things people say they do at once, even if those people vociferously disagree, and they’re all useful.
326
Why TinyML is still so hard to get excited about
This week, I went to the tinyML Summit in Burlingame, Calif. TinyML, or running small machine learning models on constrained devices, is one of the most exciting technologies I’ve encountered. But it’s also the one most likely to put people to sleep when I talk about it. Using local computing to handle object or even…
327
What I learned using private LLMs to write an undergraduate history essay
I used private and public LLMs to answer an undergraduate essay question I spent a week working on nearly 30 years ago, in an effort to see how the experience would have changed in that time. There were two rules:
328
AI and Open Source in 2023
The Highs and Lows: A Year in Review
329
Fireside Chat with Clem Delangue, CEO of Hugging Face
We discuss Hugging Face, Open Source, and AI
330
Training Deep Networks with Data Parallelism in Jax
Misha Laskin personal website. Includes a blog and projects focused on artifical intelligence.