The absolute biggest thing people are completely missing about the AI agent revolution
It’s not the days of work it saves you
It’s the years of training you no longer need to do that work
Personal Computer is coming to Windows.
Personal Computer for Windows runs on your machine and orchestrates across the apps and files you use every day.
We'll roll out first to paying Max and Enterprise Max subscribers on the waitlist.
We unlocked the working memory of LLMs 💥
Reasoning in Memory (RiM) replaces autoregressive "thinking out loud" with fixed memory blocks that form a task-specific workspace for latent reasoning.
The key idea is simple: reasoning should happen inside the LLM, not in its output!
During the dot com melt-up of 1995-2000, Nasdaq put up gains of 572%.
To date, from the 10/13/22 bear market lows, Nasdaq is up just 162%.
Listening to bears calling this a bubble top could cost you a lot of money.
Officially 1 month since I switched to a flip phone.
- Everyone is more severely addicted to their smartphones than I thought. Once you have a dumbphone, you'll frequently find yourself as the only person in the room not on their phone. It's not just teenagers, it's parents and adults of all ages. It's like everyone is stuck in a trance. 75+ year olds might be the only exception.
- All the objections I previously had for getting a dumbphone have turned out to be overblown and/or solvable. My iPhone addiction had fed my brain excuses to not do this earlier. If you really want to make the switch, you can.
- I've felt embarrassed to pull out my flip phone in public at times, for fear of being different or drawing too much attention to myself. But I have learned to just own up to it. Most people end up saying something like "Oh, I probably should do that too."
- I am using my brain more. Even though my flip phone has Waze, I find myself memorizing maps and roads. I'm more bored and get lost in my thoughts. I'm using paper and pen more. Increased desire for tangible things > digital things.
Overall, it has been a great experience and I plan on never going back.
I do like this paper.
It gives a proof that minimum neural weight norm matches minimum program length, aka Kolmogorov complexity, up to a log factor.
Weight decay work because small weights neural nets toward simpler, more compressible explanations.
https://t.co/7x8JSFjqVa
A beautiful example of an "optimal stopping problem" – Feynman's restaurant problem – with a great backstory behind it. This is a fun, well written article, and a fun math problem too.
https://t.co/0Nng9KLDHa
We’re brining local models that can run on your personal hardware inside Perplexity Computer. This will take advantage of the local hardware while giving you the privacy and token efficiency per watt, as well as access to the frontier models on the server side GPUs when necessary. Coming soon to Windows laptops. Stay tuned.
Today we're announcing that hybrid agentic inference is coming to Perplexity Computer.
Computer can split tasks between a local model running on your machine and frontier models in the cloud. This keeps private data on your device and maximizes token efficiency.
Coming soon.
My best interview in some time.
Rohin Shah leads AGI alignment/safety at DeepMind.
And he has a lot of spicy personal takes:
We probably won’t get catastrophic misalignment (00:49)
Safety 'commitments' have severe limitations (10:38)
The intelligence explosion probably isn't imminent (1:52:44)
Why he's not working to pause AI advances (51:44)
Pre-deployment evals aren't the right focus (for catastrophic risks) (37:41)
Signalling concern for safety sometimes diverts resources from actually making AI safe (01:09:51)
Reading AI thoughts is v useful for safety – and we'll probably be able to for years to come (54:17)
Governance is somewhat more likely to be the bottleneck than alignment (43:55)
Rohin's team doesn't have a veto, and that's OK (27:36)
Central banks are a promising model for regulating AI (33:34)
Also:
Google DeepMind's actual plan for building AGI safely (1:40:29)
How external researchers can positively influence big AI companies (2:21:55)
The roles GDM most needs to hire for (2:37:03)
On the 80,000 Hours Podcast. Links below - enjoy! (@rohinmshah)
Anthropic has expanded Mythos access to 150 organizations. There was pushback on this from the administration in the past, but with general release rumored to be two weeks away, there seems less point to restrictions as we get closer to public access.