prolepses.bsky.social

prolepses.bsky.social

@prolepses.bsky.social

26 followers102 following108 posts

prolepses.bsky.social reposted

Replying to @atticusgf.bsky.social

Yes - the corollary to all of this is that if you're going to show a bunch of initiative and stick your neck out, it actually has to be good or you can become a net drag.

prolepses.bsky.social reposted

"We lost control; please put us in charge of the industry” deserves a spectacularly hostile reception.

prolepses.bsky.social reposted

Replying to @zachweinersmith.bsky.social

The whole "it's just a parrot that can't make anything new" is getting extremely old when a massive massive amount of intellectual work consists of combining already known stuff.

prolepses.bsky.social reposted

DwarfStar running DeepSeek v4.1 Flash on a 128GB M5 Max at steady 15 t/s. I didn't expect with SSD streaming it could be so fast. Recent SSD streaming changes to retain the right experts surely helped, but also maybe DS4.1 uses the same experts more. Will push online when ready QA > ASAP.

Reposting with a direct link right to where @sethkarten.ai breaks down Prime Agent. The TLDR is that Prime Agent is "Jupyter notebooks for your agents", the video is absolutely worth the watch: youtu.be/n9xKblqyQ28?...

youtu.beWhy The Harness Matters More Than The Model | YC Paper ClubYouTube video by Y Combinator
Chris Patil@donotgogently.bsky.social

If you’re interested in Prime Agent by @primeintellect.bsky.social , Seth Karten (researcher at Prime, author of the Prime Agent paper) gave a nice talk on the harness at YCombinator the other day youtu.be/n9xKblqyQ28?...

prolepses.bsky.social reposted

Replying to @dferrer.bsky.social

Coming soon: AI took 18 hours and $450 to make me a personalized reimagining of Skyrim, several times what that game cost even on release. When we realize it's just not worth it?

prolepses.bsky.social reposted

Instagram's account-recovery support bot could be talked into adding your email to anyone's account, then you reset the password and you're in. If your AI agent can call a tool, assume an attacker can reach that tool directly:

prolepses.bsky.social reposted

I was reading an author recently talking about how we don't appreciate the way that email isn't siloed. That is, you don't have to be in gmail to talk to someone else in gmail. The norm is so powerful, it'd be unthinkable to change it. All email is interoperable. Social media could've been this way.

I wish more people were paying more attention to the projects that ship and have users. With the amount of change coming to our industry, spending any time listening to exposition about some experimental thing no one uses is an inexcusable waste of time. This goes double for debating skeptics.

Ed@ed3d.net

I have not been "iterating" on a thing that produces no useful output for months. Instead I have been working on @polytoken.dev while also using it for other productive work (and so have other people). It's embarrassing. It's indicative of a bad state that this is considered meaningful thought.

But are those same people running their local inference around the clock? I suspect most of them are not, and 1.7tps is about 150k free frontier tokens per day, which sounds pretty interesting to me.

prolepses.bsky.social reposted

My great weakness is that I value my own time at zero and I price goods at their component cost

i can respect that some people are getting quite far in accomplishing impressive one shots, but this is starting to look like a parlor trick in comparison to building something deliberately iterative. once you open the door to orchestration, much much bigger things are possible

In some sense, agentic programming is like employing a liar. You have to figure out how to make that liar useful even if they cannot be trusted, which is probably what it is like to design a prison or a work camp.