A festival of truth-seeking, optimization, and blogging. We'll have writing workshops, rationality classes, puzzle hunts, and thoughtful conversations across a sprawling fractal campus of nooks and whiteboards.
Ilya Sutskever and Jan Leike have resigned. They led OpenAI's alignment work. Superalignment will now be led by John Schulman, it seems. Jakub Pachocki replaced Sutskever as Chief Scientist.
Reasons are unclear (as usual when safety people leave OpenAI).
The NYT piece and others I've seen don't really have details. Archive of NYT if you want to read it anyway.
OpenAI announced Sutskever's departure in a blogpost.
Is this paper essentially implying the scaling hypothesis will converge to a perfect world model? https://arxiv.org/pdf/2405.07987
It says models trained on text modalities and image modalities both converge to the same representation with each training step. It also hypothesizes this is a brain like representation of the world. Ilya liked this paper so I’m giving it more weight. Am I reading too much into it or is it basically fully validating the scaling hypothesis?
I stayed up too late collecting way-past-deadline papers and writing report cards. When I woke up at 6, this anxious email from one of my g11 Computer Science students was already in my Inbox.
Student: Hello Mr. Carle, I hope you've slept well; I haven't.
I've been seeing a lot of new media regarding how developed AI has become in software programming, most relevantly videos about NVIDIA's new artificial intelligence software developer, Devin.
Things like these are almost disheartening for me to see as I try (and struggle) to get better at coding and developing software. It feels like I'll never use the information that I learn in your class outside of high school because I can just ask an AI to write complex programs, and it will do it...
I think that's a great answer -- assuming that's what you believe.
For me, I don't believe point 3 on the AI timelines -- I think AGI will probably be here by 2029, and could indeed arrive this year. And even if it goes well and humans maintain control and we don't get concentration-of-power issues... the software development skills your students are learning will be obsolete, along with almost all skills.
For cells to become cancerous, they must have mutations that cause uncontrolled replication and mutations that prevent that uncontrolled replication from causing apoptosis. Because cancer requires several mutations, it often begins with damage to mutation-preventing mechanisms. As such, cancers often have many mutations not required for their growth, which often cause changes to structure of some surface proteins.
The modified surface proteins of cancer cells are called "neoantigens". An approach to cancer treatment that's currently being researched is to identify some specific neoantigens of a patient's cancer, and create a personalized vaccine to cause their immune system to recognize them. Such vaccines would use either mRNA or synthetic long peptides. The steps required are as follows:
Promoted to curated: Cancer vaccines are cool. I didn't quite realize how cool they were before this post, and this post is a quite accessible intro into them.
I've been told that my number of blog posts is impressive, but my personal notes are much larger than my blog, over a million words and with higher information density. Since I've had a bit of practice taking notes, I thought I'd describe the system I developed. It's more complex than some integrated solutions, but it's powerful, modular, free, and doesn't rely on any specific service.
Most people don't take extensive notes and don't use git at all, so obviously a fancy note system isn't strictly necessary for most people. To some extent, you have to ask yourself: are your notes on some topic going to be more useful to you than a Wikipedia page or internet search or book? Do you need more records than...
Have you ever used Obsidian? Sounds similar to the method you're describing. If so, what do you think of it? Especially with respect to your preferred workflow?
A few years ago I ported my whistle synth system from my laptop to a Raspberry Pi. This was a big improvement, but I still wasn't that happy:
To get good quality audio in and out I was using a 2i2 audio interface, which is expensive, bulky, and has a lot of buttons and knobs that can be bumped.
To use a single mic for both whistle and talkbox I was using a cheap passive A/B switcher. Which feels fragile, causes pops when phantom power is on, and is one more thing to plug in.
It's hard to get super low levels of latency with the Pi. It's probably possible to get more performance out of my existing hardware than I'm managing, but as it is I'm not happy with it.
The Pi's SD card gets
Summary: Evaluations provide crucial information to determine the safety of AI systems which might be deployed or (further) developed. These development and deployment decisions have important safety consequences, and therefore they require trustworthy information. One reason why evaluation results might be untrustworthy is sandbagging, which we define as strategic underperformance on an evaluation. The strategic nature can originate from the developer (developer sandbagging) and the AI system itself (AI system sandbagging). This post is an introduction to the problem of sandbagging.
There are environmental regulations which require the reduction of harmful emissions from diesel vehicles, with the goal of protecting public health and the environment. Volkswagen struggled to meet these emissions standards while maintaining the desired performance and fuel efficiency of their diesel engines (Wikipedia). Consequently, Volkswagen...
Seems like this might work better as a control variate sort of idea, with the noise added to the environment rather than the model itself, to measure whether the model can steer its way back to its intended measured-capability level. If you add a lot of noise to an environment, then an agent which is trying to pessimize and achieve a certain lower performance will have to overperform in order to 'merely' be bad; then you can subtract the noise which you added, and estimate its performance. If it's been deliberately taking a dive, then it'll do better than ...
It's funny that this has been recently shown in a paper. I've been thinking a lot about this phenomenon regarding fields with little to no capacity for testable predictions like history.
I got very into history over the last few years, and found there was a significant advantage to being unknowledgeable that was not available to the knowledged, and it was exactly what this paper is talking about.
By not knowing anything, I could entertain multiple bizarre ideas without immediately thinking "but no, that doesn't make sense because of X." And then, each of tho...