John David Pressman
@jdp.extropian.net
1.8K followers 690 following 3.6K posts
LLM developer, alignment-accelerationist, Fedorovist ancestor simulator, Dreamtime enjoyer. All posts public domain under CC0 1.0.
Posts Media Videos Starter Packs
jdp.extropian.net
I don't think it's going to take decades to get long horizon RL. The reason you can't stack hours is that people don't know how to make general planners like MCTS work well with LLMs.
jdp.extropian.net
You can also start with intrinsically short tasks and then train your generalist agent to perform progressively longer tasks.
jdp.extropian.net
Okay but you can evaluate whether or not your action succeeded at doing something, and implicitly whether it helped you do other actions that come later. If you have a planner then you can preconmit to plans or have expectations over plans and see if actions get you in a place to do step N.
jdp.extropian.net
Who said you needed one timescale of reward? Or one source of verifiable reward? Tasks are made of parts and you can evaluate the parts.
jdp.extropian.net
Okay but what if you checked for intermediate task completion, say using unit tests you write in the moment to see if your action succeeded? And then what if you guided your search using a model trained on those boolean outcome labels and you used that as a dense reward over the other blocks?
jdp.extropian.net
There's a reason I put "real" in scare quotes but there absolutely was a rash of cases like this, the most famous being Columbine where the shooters made a doom map of the school as practice or whatever.
bsky.app/profile/keyt...
jdp.extropian.net
The default thing is that kids will ask it to do their homework and make meme edits of silly middle school in-jokes. Come on man "AI psychosis" is clearly an exceptional thing that's more like video games causing kids to shoot up schools or novels causing psychosis, "real" but extremely marginal.
Reposted by John David Pressman
keytryer.net
For the record, I think the video game panic and the AI panic are exactly alike:
- They were both based on newer technology.
- They were both claimed to be massively addictiv.
- All stories were about young people.
- All stories were anecdotal, extreme outliers, and poorly supported by data.
keytryer.net
It is republicans promoting this moral panic btw
markriedl.bsky.social
Draft bill being circulated would ban chatbots for minors www.axios.com/2025/10/14/h...
jdp.extropian.net
Okay but what does that have to do with the kinds of issues a typical normie-ish Linux user is going to encounter like e.g. configuring Wine?
jdp.extropian.net
I mean, you should in fact keep regular backups and be very wary of any command involving deleting things or "rm". It's not like Linux forums don't include trolls who will literally tell you to rm -rf things for the lulz.
jdp.extropian.net
I got access to the full text a minute ago and it seems to be a misleading headline and it's actually about not letting minors access chatbots that do porn, which could be problematic since jailbreaks aren't solved but isn't quite the same thing as a total ban no.
jdp.extropian.net
Use Ubuntu, Fedora, Mint, etc. You don't have to use weird hipster distros, just ask ChatGPT for help if you have to do stuff with the terminal.
anglesideangle.dev
how have i only seen zorin and omarchy mentioned in relation to windows 10 going eol... we are truly never going to have the year of the linux desktop
jdp.extropian.net
This is dumb, ChatGPT would have been an incredible help when I was learning to program as a teenager, would have accelerated my career by literal years.
jdp.extropian.net
I would definitely include like, some norms and thoughts about what is considered a faux pass. e.g. Using AI to write something for you and presenting it as a communication from yourself is generally considered rude. I think a lot of anger is because we haven't established norms yet.
jdp.extropian.net
Wasn't saying it was SSC, just that this is a handy link for the overall concept. All things become crab, all ideologies become Roman Christianity, which itself started off as a poor persons religion and then became kind of an inversion of Jesus's teaching.
jdp.extropian.net
I reject your metaphor and substitute my own.
jdp.extropian.net
Increasingly capable AI agents ushering in the return of solo developer projects and small teams, but I know that's a very unpopular answer.
bsky.app/profile/doll...
dollspace.gay
github.com/dollspace-ga...

It has now reached a minimal functional level to the point it is satisfied in saying it is proof that AI coding can be used to build viable products.

It will now be accepting human assistance if anyone wishes to help contribute ^_^
GitHub - dollspace-gay/PublicAppView: A third party Appview for ATproto
A third party Appview for ATproto. Contribute to dollspace-gay/PublicAppView development by creating an account on GitHub.
github.com
jdp.extropian.net
I made a Manifold Market for this.
manifoldalerts.bsky.social
Market: Will Pritzker, Newsome, or any other state governor attempt to have the state police arrest ICE agents by July 2026?

manifold.markets/JohnDavidPre...
jdp.extropian.net
So like, do you deliberately write like ChatGPT manually as a parody or do you edit your draft with ChatGPT and then edit it? If the latter I think you might benefit from a system prompt or something that makes it less ChatGPT sounding, just because people are allergic to the style.
Reposted by John David Pressman
proptermalone.bsky.social
apparently i revert back to like c. 2006 muscle memory for default website displacement activity absent social media, i've had to stop myself twice from absent-mindedly checking Andrew Sullivan's blog
jdp.extropian.net
What model is this? It looks fun.
jdp.extropian.net
Some people don't understand that there is no amount of polite argumentation that will make a clinical narcissist stop acting like a clinical narcissist.