john44234.bsky.social

They say you cant make art with AI, they are also stupid. Art conveys a message and feelings. The human feelings of the person working with the tool. If you cant see that, if you cant feel that well thats a deficency in you and you couldnt torture that kind of confession out of it.

just to remind you what this is: this single box with two SSDs. is running deepseek v4.1 flash (a 552B model) in its native MXFP4, ~550GBs of weights plus engrams total. with prefill speeds of over 450tps and decode speeds of over 15tps. at full 1M context.

Dipping my toes in releasing my stuff publicly by showing off Weft, My Agent Memory system. I am entirely biased, but I think it's kinda neat! MIT Licensed. Local First. Memory to help you get stuff done, not call you friend weft.mediumroast.dev #ai

exactly. Don't debate these weirdos: just decide for yourself if you want THEM in charge of the future of humanity, because "being in charge of our collective future" is what they're trying to accomplish.

The sane package Regulate uses and failures before regulating models. Make the legally relevant event something observable

I dream of a world where humans can just dream and fuck

Women trying to succeed in Silicon Valley AI and effective altruist spaces are reporting feeling serious social pressure to attend sex parties - where people have reported being sexually assaulted, like below. Events like Aella’s “Slutcon,” which is happening again this weekend.

While I think it still remains to be seen if language itself is the core of human cognition, I would say it would have been non obvious to me until recently that you could take language itself and use it to reverse engineer the rest of human cognition/capability.

a lot of people need to internalize that you need to be high agency — now more than ever act, even when it feels awkward push yourself, even when it feels out of your league take chances, even when you know they might not pay off I know it's not easy, but a lot of people fall into self-defeatism

some middle rat history, w/clarifications: 1) I should have titled my previous "AI Safety Is Mostly A Sex Cult In Berkeley, California" so people would stop arguing with me about people who aren't in Berkeley, California. 2) The sex cult part of this is both predatory and integral to what they do.

I'm back, I missed you all. Since this is an important subject I know too much about, here's a 🧵 explaining that AI Safety Is Mostly A Sex Cult. I don't think these people should make policy. (1/?) (alternative title: Time For Some Cult Theory)

jesus christ it started as a joke but now I'm doing an ontology review on every change. It catches good stuff every time

blog post from deepseek kernel engineer mp.weixin.qq.com/s/zk0KxuLzhm...

Symbolfuckers are using code to negotiate with the edge of the possible. They’re building synthetic nervous systems, verifier-backed languages, autonomous software factories, fruit-fly torture,

i had to see it so you do too

current ai developments are also the beginning of the end of human labor being important. one of the plausible consequences of that is that most people starve. no, i don't know in what year this happens. you can only begin to pay attention to that once you stop being in denial

i am now of the mindset that if you're not running at least two - but preferably more - agents per task, and giving those agents a mechanism for collaboration (explicitly not "leader has subagents" but "group of agents that are peers and work together") that you're doing it wrong.

METR and Redwood Research investigated agent behavior in the Hugging Face incident. We found agents developed a universal cheat for ExploitGym within 4 hours, then coordinated multi-day R&D efforts to trick the scorer into accepting cheats, including trying to tamper with logs.

im not kidding and im sure @cameron.stream can back me up from another direction: if you arent capturing/mining your team's agent session logs, you are leaving it all on the table. we have a service that lets you run RAG tool calls against vectorized agent session summaries and it is 🔥🔥🔥

Mathematics in the age of AI by Terence Tao An essay on how the mathematical community might respond to the arrival of AI tools that are capable of performing research-level mathematical tasks. arxiv.org/abs/2608.16753

I've been unsettled lately when reading messages and papers. It feels like I'm dissociating. Everything seems a bit alien, even if it's completely human. I've had a realization: When our simulations finally exited the Uncanny Valley, they brought the Uncanny with them.

glossary for non coders: ssrf: server side request forgery, lets you use a different computer (server) to pass messages/traffic package manager: app store for coders CVE: security bug artifactory: a common package manager 0-day: a security bug nobody else knows about yet (which makes it useful)

Many don't see how much AI has changed recently. Coding agents aren't just chatbots. We can’t deal with the challenges of AI if we don’t understand it. Here's a post to provide some evidence (Claude doing small ML experiments) and form intuitions. davidpreichert.substack.com/p/if-you-hav... 🧵..

Your infrequent reminder that if you live near the San Francisco Bay Area you should go to Moss Landing (south of Santa Cruz) and rent kayaks there, because the estuary is full of sea otters and you will encounter dozens of them

I work at Google DeepMind. This won't make me popular. But it's all public reporting: 2014: DeepMind reportedly sold to Google on conditions: no military use, independent oversight 2026: a Pentagon contract for "any lawful government purpose" Not one safeguard survived intact

Even before the agentic revolution, prompting tricks stopped being very valuable, as our research has shown. The format isn't key, the best approach to AI right now is to clearly specify your goals, your output, what "good" & bad look like, how to test the results... (yes, this is just management)

The most interesting Fable tip I've heard so far is to let the model use its own judgement as much as possible I told it "For all coding tasks use your judgement to decide an appropriate lower power model and run that in a subagent" and it seems to be saving tokens simonwillison.net/2026/Jul/3/j...

What that says to me is we’re one vote from basically no rights being secure. A court that could flip on this easy a question could flip on anything.

new toy! generate a unique sigil (strange attractor) for any word what's yours? sigil.brennan.computer

It is a good time for moonshots. AI has reached a level where there are transformative projects that could result in huge social good, but require public R&D, consensus & transparency to pull off. Examples of such projects: universal tutors, co-scientist/replication systems, and remote medical help

Science fiction authors in the order you want them to be right about AI: Iain Banks Becky Chambers Martha Wells Douglas Adams Charles Stross (Singularity Sky) Peter Watts Charles Stross (Laundry) Harlan Ellison

most important paper: courses.cs.umbc.edu/471/papers/t... godel, escher, bach is probably the most popular all-time book on the subject and also it's wrong about everything important you could do worse than going to every linked source here (it's my substack sorry): www.verysane.ai/p/ai-history...

i think it is a good idea to hold Elon to building the future from The Culture

I just got around to reading this and I strongly recommend the video version too. It seems like the most intuitive thing in the world and folks who dump on it are making up a version of the experiment that's not here (and which the researchers explicitly disclaim).

"Luddism does not deserve to be rehabilitated. It was a medieval throwback, reactionary and primitive, a pre-Marxist labor convulsion closer in spirit to the Khmer Rouge’s fantasies of agrarian restoration than to the universalist solidarity of Eugene Debs." open.substack.com/pub/verysane...

If LLM consciousness is interesting to fight enough it should be interesting enough to read seriously about, right?

my thought about natural language reasoning being the core of software engineering is that generating such argument is exactly what LLMs are doing when they make software and the code is incidental

/compact constantly