Open Source AI Art w/ @banodoco | 👫🐕 w/ @hannahsubmarine
Hallihallo / Howiya
Joined December 2011
- Tweets12.1K
- Following2.5K
- Followers15.7K
- Likes21.4K
I wish they kept going for 4-5 more pints with John pressing him on this point until he’d eventually confess that - deep down - he believes that the absolute best way to distribute tokens would be through one central org over which he’d be the wise and firm authority
sam: local inference is a terrible allocation of scarce resources
“a lot of people want local inference, and somehow they assume that the problem is solved now because we have open-source models, we're going to be able to run them on our computers, and so we're not going to need all the stuff in the data centre
but from a global allocation perspective, actually it's quite terrible to have everybody, like, have a bunch of the world's limited supply of DRAM sitting in their own computer and being used five percent of the time
you want maximum utilisation in the cloud if you're trying to deal with this very supply-constrained environment
somehow many people's intuition is the opposite of that”
2 months after Fal said they're release H3 Max, it remains unreleased
Meanwhile, Pika took the opportunity to release a H3 fine-tune that performs better than it:
I wasn't a fan of the entire GPT-6+ line but Terra 6.1 feels like a step back in the right direction!
Though inferior in terms of raw intelligence, its intuition feels vastly improved on Astra. Far less annoying to work with.
Makes me very optimistic for Luna/Sol/Astra 6.2!
(though if it weren't for subsidisation I would be using Deepseek 4.1 Flash for almost everything I do w/ OAI models rn)
Every time there’s a new Chinese model release after some meaningful event - Xi visit to US - and new round of of fear-mongering, I breathe a sigh of relief 🧘♂️
🚀 DeepSeek-V4.1-Flash-1017 is here!
Stronger benchmark performance, faster task completion, and shorter outputs than the previous version—a Pareto improvement in performance and cost. Now live on our website, app, and API.
🛠️ Optimized for @DeepSeekHarness, Tencent CodeBuddy, WorkBuddy, and more. Upgrade to DeepSeek Harness v0.2.1-rc.1 or later for the best experience with the new model.
1/4
I just wish it was back near the OG price
Luna/Haiku are far inferior but the subsidies make them difficult to beat rn, around 98% of my tokens go to smolbois
POM retweeted
.@willob of @ulyssesinc joins @sacca and me for a pint to discuss how subsea robotics is unlocking Earth’s last frontier. We talk about the physical constraints of operating 1,500 meters underwater, why seagrass holds 20% of the ocean’s carbon, how green tape hinders environmental restoration, and why hard tech startups are verticalising hardware manufacturing. (Plus, Will's interesting Guinness history facts.)
If anyone builds it, everyone thrives!
interpretability might be the most important scientific problem in the world right now, as I don't think we can solve alignment if we don't know what we're building. I think I have good news about it. I co-authored early work in mechanistic interpretability at OpenAI and came away pessimistic, because neural networks appeared to fold into arbitrary tangled knots. years later, I started openai's first large-scale automated interpretability project, finding about 2% of neurons legible with the primitive tools we had
with modern agentic tools and a much closer look, the picture looks totally different. the vast majority in models up to trillions of parameters are legible, and their circuits beautiful. many famous examples of tangled neurons turned out to be misunderstood. The roadblocks I expected mostly aren't there. I now think ambitious interpretability will be solved by the time we need it most, and possibly quite soon, as automated scientists keep improving. it may be a bitter lesson for interpretability: the simple thing, at absurd scale, works
already-clean.goodfire.com/
Operate under the assumption that every piece of information you create that's digitally accessible may be some day be public
Every message you've sent, your internet history, etc. etc.
Make peace w/ it. Ether lose your shame or don't do and say things you'll find shameful
I asked Astra to attempt a rough but evidence-based estimate of the number of people who make a living from playing chess since Deep Blue defeated Karsparov:
Imagine how absurd Kasparov would've found the idea that there's be almost 1000 people employed in a computer chess company or a cohort of hundreds of professional chess socialites
Gossip Goblin is clearly extremely impressive as an artist but his representation is perhaps equally impressive at what it does - kudos, @fableshowrunner!
AI Film ‘Gods Don’t Give Gifts’ Sets First Trailer for December Theatrical Run, Plans Oscar Bid for Best Animated Feature variety.com/2026/film/news/a…
I was probably wrong in my indignation here: