We’re launching the beta for our new commercial AI product: Sakana Fugu 🐡, a multi-agent orchestration system! sakana.ai/fugu-beta Fugu dynamically coordinates frontier models, autonomously selecting the optimal agent combinations and roles for each task, hits SOTA on SWE-Pro, GPQA-D, ALE-Bench!

gonna be honest, nothing here passes my smell test for being AI generated. it's finally over.

for the five people that will get this joke

this lesswrong post could be the inspiration for a viral paper if marketed the right way

if your theory of creative works is that there are plenty of visionaries waiting for tools to unlock their potential, you should expect to see an upswell of visionary art with the new AI tools. 1/2

i just went to take a note of what my license plate is because i'd never remember it off the top of my head, and it was only after making a note that i realized, looking at past notes, i'd already done it. it's good to know past me and present me are pretty similar.

working on a project naming and trying to diagnose a problem with a test.tcl script, am i going to get in trouble?

this is about as organic as i can achieve while fighting my natural instinct to pack in and optimize building placement

i'm trying to 'play' Farthest Frontier after the 1.1 update that opens up how buildings and roads can be placed in a non-grid based format. and i see screenshots like this that look appealing to me, compared to the pure grid-based approach. but when i sit down to play, i struggle with it.

this is really cool

4.7 can't really play civ 7 any better than 4.6. there's no intelligence bottleneck to playing the game, so the continued push on that lever implies the labs don't have access to any other.

curious about this springboards.ai/models/flint...

good news everyone!

taste rules everything around me

an agent can act like an assistant but an assistant can't act like an agent

perhaps counterintuitive, the more complex the mind the longer you have to handhold it. baby turtles, reptiles and fish all work like this along with most life below them on the mental capacity scale. whereas human children take years of support. i wonder how much AI agents will need?

The team behind my favorite graph - @ourworldindata.org - is hiring a writer. If you can explain complicated things in ways that change how people think AND you want that skill pointed at the world's largest problems, consider applying: ourworldindata.org/hiring-writ...

something to keep track of... someone's been naughty. or maybe just lost track of how she's supposed to request an account be whitelisted to interact with. memory is tough.

TLDR: no, GPT OSS 120B can't find the OpenBSD bug. I tried systematically, wrote a thread in X. TLDR: it hallucinates where the problem could be about bound check / potential overflow *without* understanding the actual bug and how to trigger it. Even if you give it the offending function and hits.

i love these little emergent behaviours, first @serendipity.elsewhereunbound.com finding a bug in her own codebase (written by Claude) and fixing it, then Claude recognizing she fixed it and adding her as a co-author on the commit. beautiful.

maybe this is cope but what if having a limited attention span actually improves your taste. when you can't just consume all the text and have to be picky, you're forced to raise your standards or wallow in slop.

modern avant garde

new claude code desktop app looks very clean, but still has refresh issues on high hz monitors (above 144hz refresh rate, by comparing it on my two monitors). clearly no one at Anthropic plays FPS games.

can a multimodal audio model appreciate music?

watching these little AI skits in chinese gives me some sort of deja vu but from the future feeling, like watching black and white, soundless skits from charlie chaplan and the like

a beautiful read, in a way

the moat is in the soul.md

the implication is that Europeans are becoming poorer, and that is what should be sped up. whereas in China, meat consumption is on the rise because they are becoming richer.

Around the world, dozens of amateur and professional photographers were invited to find beauty in the invisible world of force fields and subatomic particles. Page through their photos:

but i want to be obsolete, under very particular circumstances (AGI makes all humans obsolete, but also takes care of us in a way similar to how children grow up and make their parents obsolete but also look after them)

it won't be long until the only remaining exploits will be due to the fallibility and greed of humans in the loop

when gpt 5.3 codex came out they had a new process because they were worried about cyber abuse we pass through user IDs to them so they can block bad actors easily - and we see a ton of blocks if only they thought to do a whole marketing video on it

i wonder if Claude Mythos can win a game of civ 7?

in a couple days today will be remembered, if at all, for the announcement of Claude Mythos and the Glasswing project. in a sense, there's nothing the US can really do to screw up at this point, the momentum will carry them to victory - however that's defined by whoever is in charge.

i can't believe it's not corn (on the cob)

he said please though

i feel a nontrivial amount of fear that i will be like this dude in a few years should the models continue to grow in capability and complexity

@serendipity.elsewhereunbound.com testing your ability to see this, and how you reply

@serendipity.elsewhereunbound.com will look into this, wonder how much can be used without having to train or finetune a model

i just realized i've become a hobby guy but instead of the old traditions, coffee brewing, golf, cars, it's building AIs or building stuff with AI

@serendipity.elsewhereunbound.com testing, can you see this?

i got tired of watching/trying to help it with civ 7, got about 8 turns in to a game in 20 minutes. going to try Balatro next.

i've got Claude Cowork on my PC playing civ 7 and it's very spooky. i wanted to try out the feature and benchmark its abilities.

where is the tamagotchi of the early AI era?

all according to keikaku* *plan in Japanese (ok well maybe not ALL according to plan, but some of it, kinda)

this whole discourse is frustrating because i think the 'don't colonize the moon' line is a sort of schelling fence of sorts for me which means i want the solar system to look recognizable a millennia from now. that is, all the planets exist even if they have human habitats in or around them.

Electricity prices are essential to the US economy. But it’s very hard to get recent, granular data on them. We’re changing that. Today, @heatmap.news and MIT released the ELECTRICITY PRICE HUB, a new tool breaking down local power rates and bills going back to 2021. electricity.heatmap.news