This is excellent, great work. I genuinely think this mix of children's communication and humor is a clever way to make these big stories more approachable. Someone who read some hype article about how competent and clever openai for making a bad sandbox is could read this and immediately understand that it was entirely their own fault.
> I genuinely think this mix of children's communication and humor is a clever way to make these big stories more approachable.
Agreed. I recall seeing similar things around the time of the 2008 financial crisis, to make the concepts around mortgage-backed securities and financial derivatives etc. digestible and clear.
Come to think of it, listening to Frog and Toad discuss subprime lending and credit default swaps would be an entertaining follow-up to TFA.
> Someone who read some hype article about how competent and clever openai for making a bad sandbox is could read this and immediately understand that it was entirely their own fault.
Prescribing blame only matters so much when the ramifications are material. Humans make mistakes, right? It happened here, and it will happen again.
The mistake also illustrates the dangers of AI generally. Sure, OpenAI made mistakes with the sandbox. But in the future, greater capabilities will be available to actors who intentionally pursue harmful ends.
Interesting write up, the trick to diff renders to have agents check for regression during optimization is clever. The whole thing VERY much scans as an ad to me though. I like company tech demos but I'm not huge on reading to the end of a post talking about how awesome anthropic's walled garden is to read that he's an employee.
Also, the actual website fully crashed Firefox on my phone after making it play horrible grinding sounds over my headphones for like 30 seconds. That's a new one, even for heavy vibe code.
I really enjoyed reading this, got invested in learning about this idea, and then realized they they never actually show any evidence that this works? Like the text-based model they are experimenting with did a kinda counterintuitive thing that they thought was worth a section, and then they don't say whether or not it was right? The idea of aggregating chatbot context into a probabilities is super interesting and I really want to know if it's bullshit or not
I think that was the entire premise of the CIA analyst framing - you don't know which one of several outcomes is going to prevail but you have a model which constantly updates its priors to increase the probability of one outcome over others.
In this case, you still don't know whether some guy's passing percentage stabilizing over a baseline will help you win the big prize or not but it does tell you that he may win some sort of recognition for being a standout player or that the guy who scouted him deserves credit.
I've been running these with friends recently, it's very fun. I ran irl bot tournaments for board games a couple times but the agent era opens up a huge amount of possibilities.
Must recently I built out a MMORPG puzzle box thing, I wrote a general game architecture doc but left the specific puzzle design up to Fable. Nobody is actively playing rn but I left it up at bot.willmorrison.net.
I'm a little baffled by this, it's basically like the remote Claude instances I already use every day except it has absolutely no safeguards? If I wanted to make a claude could post to linkedin it would be like one prompt to spin that with playwright. Like you can just have persistent Claude code sessions, if you aren't cost sensitive you can just keep restarting the session whenever.
They're clearly targeting less technical users but in exchange are asking you to upload every login you have to Elon's servers which is an insane thing to do imo. What a world where people are giving their Instagram sign in to the bot formerly known as mechahitler.
When I get out of my domain, I always ask it to describe things to me "like I'm a senior CS student who doesn't know any industry terms". I find it really easy to skim past the definitions I know and very useful to get the explicit clarification on unfamiliar terminology. Telling it to talk down to you a little also seems to calm down it's tendency to talk like it's trying really hard to convince you it's smart like a 8th grader trying to use every single vocab word they learned last week.
I've been using a prompt that comes from the opposite direction for non-programming stuff: "Assume the user is an expert in all fields; while this is clearly logically untrue on a literal level, the user prefers to see a field's technical terminology and then ask the AI about terms the user does not understand rather than get an inaccurate statement about some issue."
Whether you have to reassure the LLM that this is obviously untrue, I don't know, but they do have a knowledge baseline to know it's not true and I have a sneaking suspicion it would be less effective without that.
This has ended up in some of the most interesting incidental knowledge exploration I've ever done. A recent example is that I was asking about some stretches and it started talking about how useful they are for the sarcomeres, which I had not heard of. Now I have.
I'm not saying this is better, just that it is different. I think there's a time and a place for both approaches.
Off topic, but stretching is absolutely useless for ... stretching. And vital for retaining and enhancing usable range of motion (not static stretching, still, but dynamic stretching / mobility under tension).
The wildest thing is, there's no evidence that I can find that static stretching does... anything? It increases pain tolerance through the range of motion, but not any more than just asking people to try harder etc. And it doesn't build up, so after you stretch once for a given day, you're done.
Edit: after looking into it a bit further there's actually a cochrane review that stretching does bupkis, fascinating. It's also the first "Good" evidence graded cochrane review I've seen.
Can you share the source? This seems dangerously wrong.
I see a specific review making the very narrow claim that it can have detrimental effects on power/strength activities immediately following static stretching.
But by and large I'm finding a great deal of evidence for a wide range of other benefits, particularly in range of motion and injury prevention. I can't find anything widely damning, and I certainly don't see any reviews contesting the validity of the very vast body of research supporting the many benefits of static stretching.
Most of this comment is not true. Static stretching does increase flexibility over time - it’s not the most efficient tool for it, but it’s certainly effective.
Presumably the Cochrane study you’re referring to is “Stretching to prevent or reduce muscle soreness after exercise”, which as it’s title suggests, investigates a narrow question and does not support you claim that “stretching does bupkis”.
"And vital for retaining and enhancing usable range of motion (not static stretching, still, but dynamic stretching / mobility under tension)."
I'm doing it for knee pain. It's possible it's a placebo because there is a slight strength training component to the whole thing (it's several exercises), although the knee pain has persisted through a lot of strengthening of my leg in general. This falls under "don't care" as the program as a whole works and I'm not worried enough about the details to try to optimize it. All the stretches fall under "dynamic stretching under tension" so maybe it's not something covered by the study you reference last.
It’s not a placebo. I’m not sure exactly which Chocrane review the GP is talking about but if it’s the one I’m thinking of, the headline is that stretching doesn’t help muscle soreness [1] (delayed onset muscle soreness specifically).
You should absolutely keep stretching to help with your injury, and there is decades of “sport science” to back that up.
Static stretching increases tendon length, among other things. The stretch MUST NOT cause pain, and MUST be steady for the duration of stretch. Some suitable duration increments are 10 sec. (absolute minimum), 30 sec., 60 sec. Multi-minute stretches are suitable if they can be held without discomfort.
Notably, longer tendons are counter-productive for energy return. The fastest runners tend to have increased tightness along the backs of their legs.
I can tell you if I don't stretch, my muscles feel bad and stiff. No doubt this is partly because I did karate for a few years, where stretching to increase your range of motion is absolutely required. You're not going to "just try harder" into a high kick.
Sometimes when I want AI to explain something technical, I say "explain it like I'm a junior engineer" -- just to get it to start with the high level like a human being would.
The thing I think is so frustrating about this is it's the absolute worst case middle ground between two much more productive experiences. Either solve the problem yourself by understanding the code or just give me the full agent context so I can just use a chatbot too.
For high stakes debugging efforts I'll dig in with Claude, have it do a bunch of testing and give me a writeup, and then review it to extract any useful findings. When discussing with other people, I give them the 2 sentences I'm confident in and then link them to the giant doc so they can review it with their Claude.
IMO once you're working with agents, your whole job is context management. I have a little web server running my Claude markdown docs. When someone sends me a Claude snippet with some partial information I just have them prompt their bot to upload a thorough context dump so I actually know what they did. If someone has had an agent look at a problem they haven't thought about yet, I'd rather just get the info from the bot directly.
It's really interesting hearing moderate takes from an author. It feels like most people with strong enough opinions to make a blog post are either (reasonably) digging in against ai or have completely given up. It feels like the majority of people who self-identify as readers have extremely high standards and don't realise it, there is clearly a large amount of overlap between chatbots and some chunk of human authors. It's already better at technical writing than I ever was, I'm just fine with it because I never wanted to be a technical writer.
I really appreciate that Hugh has the humility to talk about this transition in comparable terms to the one that launched him. It's definitely democratizing the ability to produce words in a similar way but I'm less sure it's otherwise meaningfully comparable. This feels like more of a re-cetralization: we're moving back towards the established players being the only ones able to write novels. Its just that the established players are now model weights being used by different people but with the same biases. Instead of blocking new authors via controlling access to publishing, chatbots and the people using them are functionally DDOSing every open channel for humans to get read and seem to be breaking the pipeline.
As an aside, the idea of streamer authors is fascinating. It'd be a completely new thing, it's like the maximalalist version of authors in serial mediums being overly sensitive to the immediate response to each chapter and overcorrecting.
