anishathalye

6 ms·

1.
Show HN: Gandalf the Grader(github.com)
1 points by anishathalye 4mo ago|0 comments
2.
by anishathalye 4mo ago
I recently gave a guest lecture on AI Agent Security in MIT's 6.566 class (a survey course on computer systems security). We covered: - Foundations of LLMs, from next-token prediction to conversational chat and tool use - Foundations o
3.
AI Agent Security Lecture(github.com)
4 points by anishathalye 4mo ago|3 comments
4.
Mypy 2.0 Relased(mypy-lang.blogspot.com)
4 points by anishathalye 4mo ago|0 comments
5.
by anishathalye 6mo ago
I was working on AI in 2018 too :) At that time, I was quite interested in adversarial examples and ML security.
6.
by anishathalye 7mo ago
MIT already has an excellent class on Ethics for Engineers: https://e4e.mit.edu/
7.
The Missing Semester of Your CS Education – Revised for 2026(missing.csail.mit.edu)
457 points by anishathalye 7mo ago|132 comments
8.
by anishathalye 7mo ago
We (@anishathalye, @jjgo, @jonhoo) returned to MIT during IAP (January term) 2026 to teach a new iteration of Missing Semester, a class covering topics that are missing from the standard computer science curriculum. Over the years, the thre
9.
The Missing Semester of Your CS Education (2026)(missing.csail.mit.edu)
9 points by anishathalye 7mo ago|1 comments
10.
by anishathalye 1y ago
Good point, I see how the example can be confusing. Updated the example to have `reverse=True` and a comment, hopefully that clarifies things.
11.
by anishathalye 1y ago
For sure! I share your feelings about good science and honesty in academia :)
12.
by anishathalye 1y ago
Yeah, that was just a design choice that I made: I wanted a library that worked with `Iterator`s, felt more lightweight to me / fit my immediate needs better. I'm personally not a huge fan of Pandas DataFrames for certain applicat
13.
by anishathalye 1y ago
I think semantic data processing in this style has a nonempty set of use cases (e.g., I find the fuzzy sorting of arXiv papers to be useful, I find the examples in the docs representative of some real-world tasks where this style of data pr
14.
by anishathalye 1y ago
That was a small self-contained example that fit above the fold in the README (and fwiw even last year’s models like GPT-4o give the right output there). That `sort` is based on pairwise comparisons, which is one of the best ways you can do
15.
Show HN: Semlib – Semantic Data Processing(github.com)
63 points by anishathalye 1y ago|12 comments
16.
by anishathalye 1y ago
Hi HN! I've been thinking a lot about semantic data processing recently. A lot of the attention in AI has been on agents and chatbots (e.g., Claude Code or Claude Desktop), and I think semantic data processing is not well-served by suc
17.
Semlib: LLM-Powered Data Processing(anishathalye.com)
1 points by anishathalye 1y ago|1 comments
18.
Linearizability testing S2 with deterministic simulation(s2.dev)
7 points by anishathalye 1y ago|1 comments
19.
by anishathalye 1y ago
Looks like the 2023 lectures weren't uploaded to YouTube, but the lectures from earlier iterations of the class, including 2022, are available publicly. For example, see the YouTube links on https://css.csail.mit.edu/6.
20.
by anishathalye 1y ago
Implementing an ACME client is part of the final lab assignment for MIT’s security class: https://css.csail.mit.edu/6.858/2023/labs/lab5.html
21.
Sleep-Time Compute: Beyond Inference Scaling at Test-Time(arxiv.org)
1 points by anishathalye 1y ago|0 comments
22.
Co-Adapting Human Interfaces and LMs(jessylin.com)
46 points by anishathalye 2y ago|4 comments
23.
by anishathalye 2y ago
I see someone posted this before I was able to do it :) Hi HN! For the last six years, I've been working on techniques to build high-assurance systems using formal verification, with a focus on eliminating side-channel leakage. I'
24.
Overcoming Hallucinations with the Trustworthy Language Model(cleanlab.ai)
2 points by anishathalye 2y ago|0 comments
25.
by anishathalye 3y ago
Very cool! That's a nice design that gives the developer the choice on the trade-off between being upgradeable and being future-proof against developer key compromise. Transparency logs indeed are a neat ingredient to use here. I'
26.
by anishathalye 3y ago
This is really neat! We've been working on some research to formally verify the hardware/software of such devices [1, 2]. Neat how there are so many shared ideas: we also use a PicoRV32, run on an iCE40 FPGA, use UART for communic
27.
by anishathalye 3y ago
Sharing some context here: in grad school, I spent months writing custom data analysis code and training ML models to find errors in large-scale datasets like ImageNet, work that eventually resulted in this paper ( https://arxiv.o
28.
Automated Data Quality at Scale(cleanlab.ai)
1 points by anishathalye 3y ago|1 comments
29.
Fine-Tuning OpenAI Language Models with Noisily Labeled Data(kdnuggets.com)
2 points by anishathalye 3y ago|0 comments
30.
Compromising Garmin Sport Watches: A Deep Dive into GarminOS and Its MonkeyC VM(anvilsecure.com)
176 points by anishathalye 3y ago|74 comments

More ›