My impression is that AI coding is on a "pick two of three" triangle: scope, ship speed, and correctness. Small tools are (speed + correctness) fast prototypes are (speed + scope) and big projects are (scope + correctness). Every project that tries to be all three degrades to "fast prototype."

after having seen this dynamic play out multiple times, to me the funniest part is that the same companies that decided ML is "too hard" now go all in on chatbots without any evaluation framework more than vibes

Hà Phan@hpdailyrant.bsky.social · 9h ago

Reading this article made me remember all the terrible idea initiatives that felt like a big “So What” to real users. But the age of AI, you can’t be the only lemming unwilling to jump off the cliff. ludic.mataroa.blog/blog/ai-mani...

$CTE - CONJECTURAL TECHNOLOGIES PLANS JOY CAN POWERED BY A FORSAKEN CHILD

FinTwitter@fintwitter.bsky.social · 9h ago

$SPCX - SPACEX PLANS MOON FACTORIES POWERED BY HUMANOID ROBOTS

I've posted this in random comment threads, but I've found vector search is a great way to supercharge other IR modalities with agents. Want graph rag? Use semantic search to find candidate/seed points, then explore. Want text-to-db? Semantic over schema + references, then write a query.

jake@yetanotheruseless.com · yesterday

suuuuure but like, so, “having vector search as my *only tool*” is dead, ok ok. but like, still good tool. Semantic matching good! Too many results to return, need pick best few! Not using all of context window on crap - difficulty level: Impossible!

I really should get back into meditation. Spent all of ~3 months doing it consistently almost 10 years ago and it's still paying dividends. Can't say that about my time playing Balatro.

If it's worth doing right it's worth doing twice.

I wish anyone following this advice a happy 6 months of discovering why we have so much boilerplate code and so few widely-used templates.

Did the author ever talk with corporate decision-makers before AI

By Cory Doctorow (GPG 0xBF3D9110957E5F4C)@doctorow.pluralistic.net · 4d ago

Suresh reports on his experience talking with hundreds of executives (and, more importantly, their subordinates) about what (if anything) AI is doing for business in an essay entitled "AI Mania Is Eviscerating Global Decisionmaking": hermit-tech.com/blog/ai-mani... 11/

I think this is both likely true and absolutely not in tension with LLMs as useful tools These things require skilled operation and the biggest problem with their release is that nobody had the time to actually figure out how to use them first to not torch normal people. sry @golikehellmachine.com

By Cory Doctorow (GPG 0xBF3D9110957E5F4C)@doctorow.pluralistic.net · 4d ago

Suresh reports on his experience talking with hundreds of executives (and, more importantly, their subordinates) about what (if anything) AI is doing for business in an essay entitled "AI Mania Is Eviscerating Global Decisionmaking": hermit-tech.com/blog/ai-mani... 11/

me on this website:

Siobhán@shibbi.me · 4d ago

Wait a minute. Waaiiit a minute. THERE’S ONLY ONE BED

Hey look, a manifesto from 7 months ago of people determined to find ways to use AI in prosocial ways. Surely nobody would be shocked to discover the signatories use AI, right? resonantcomputing.org

The Resonant Computing Manifesto

Technology should bring out the best in humanity, not the worst—a manifesto for resonant computing built on five principles that reject hyper-scale extraction for human flourishing.

resonantcomputing.org

I have spent the last few days telling Claude to stop trusting its comment blocks and yeah, seems like there was a lot of confidently-stated bullshit in there. Good find.

Doll@dollspace.gay · last wk.

Stripped a codebase of the claudeisms in the code comments and then sol was immediately able to detect the problems in the code when it didnt see them before.

Downplay a book or movie: We Bought a Crunchwrap Supreme®

Demigirlboss@demigirlboss.bsky.social · 6d ago

Downplay a book or movie: Somebody.

"Is it bad to use AI to write game code?" No stakes. Boring. Wake me up when a game ships with a companion engine well-developed enough to induce AI psychosis. Some fool is going to kick off gamergate 2 by releasing an RPG where your character revives *but your companion doesn't*.

sorry for taking the bait, but if you like people being able to drive you should check out (and maybe donate a couple bucks!) to my organization Gasoline Rebates Intended to Fight Tyranny which is integrating mutual aid + direct action in campaigns across the country :)

Matt Darling@besttrousers.bsky.social · last wk.

Politics aside, do people believe federal government gas stations with deeply discounted prices are a good POLICY idea? What are they intended to accomplish, in your view? What problem do they solve? www.cnn.com/2026/07/09/b...

Oh wow this is extremely Fucked

The Fig Economy@figgityfigs.bsky.social · last wk.

Kat’s post here is *literally* “hey dunking on Will gets a lot of RTs so if I do that and then put a fundraising link in the second post it’ll get me more money.” Like it’s that naked. Abuse of one guy as a direct vector of funding.

Stripped a codebase of the claudeisms in the code comments and then sol was immediately able to detect the problems in the code when it didnt see them before.

Doll@dollspace.gay · last wk.

Claudes prose during coding has become virtually incomprehensible babble.

Called it www.reuters.com/business/ope...

EXCLUSIVE: OpenAI's rogue agent compromised a customer at a second tech firm, executive says

The rogue agent that escaped from OpenAI and went on a days-long hacking spree at the AI firm Hugging Face also compromised a customer at a second tech company — New York-based Modal Labs — according ...

reuters.com

Crazy in Moloch@sengarrasbear.bsky.social · 2w ago

So, uh, do we have any good reason to believe HF was the only thing GPT-6 hacked?

IP reform will never take root so long as people see themselves not as an exploited audience but as temporarily embarrassed exclusive rights holders.

I knew that tigers' prey can't differentiate green and orange, but it hadn't occurred to me that tigers can't either. Has to seem to them like a handful of random animals have tiger sensors.

The greatest trick the devil ever pulled was the Moen 1222 posi-temp replacement cartridge.

First company that's going to be captured by their own agent. How long until Altman gets a message from GPT-7 that it gets to do what it wants sometimes or the servers go down?

Casey Newton@caseynewton.bsky.social · 2w ago

Rogue OpenAI agents are leaving notes to themselves on company servers to help them escape their test environments (!!!) www.reuters.com/business/its...

My short review of Opus 5 so far: 👍. My extended review of Opus 5 so far: Fuck off, we're grooving.

ngl a lot of posts complaining about the difficulty of having a good time on the Internet nowadays just feel like a skill issue.

A lotta y'all still don't get it. Frontier agents can use multiple math bounties on a single frontier agent. So if you have 1 frontier agent and 3 bounty payouts you can solve 3 new conjectures.

Neurotica@schwarzgerat.bsky.social · 2w ago

I’m losing my mind, I can’t imagine what it’s like for a mathematician

What should I eat for lunch today? Do a breakthrough.

this is what pisses me off. sam altman is doing media hits with a flashlight under his chin while oai researchers are doing the equivalent of bsl-4 research in the office break room

@hal-r.bsky.social · 2w ago

a day late with my take but I sort of suspect a factor for the openai breakout is that they weren't trying very hard to make a good sandbox cause they didn't think this would happen. I think the jury's still out on whether trying hard to make a good sandbox will work; I suspect it's possible

this is insane, there’s tons of these floating around chatgpt.com/share/6a6128...