just had a very productive call with dijkstra. goto is back on the table

Politics is defining the enemy. Everything else that politics claims to deal with becomes something mundane — non political — in the absence of a quest for an enemy. Look at the contortions polities without natural predators in objective reality go to invent themselves some enemies.

neutral@neutral.zone · 6h ago

The problem with politics seems to be the people interested in politics

Life hack: read the 1 and 2 star reviews. If it's mostly "delivery dude was rude" or "this hammer is useless to remove nail polish, avoid" the product's good.

Max Dubler 🏳️‍🌈@maxdubler.com · 3d ago

The amount of reading you have to do on Reddit to background check any purchase over $50 is wild. Side note: you absolutely cannot trust Amazon reviews because sellers routinely give away free product and cash in exchange for five star reviews.

AI is just next-world prediction

Life hack: read the 1 and 2 star reviews. If it's mostly "delivery dude was rude" or "this hammer is useless to remove nail polish, avoid" the product's good.

Max Dubler 🏳️‍🌈@maxdubler.com · 3d ago

The amount of reading you have to do on Reddit to background check any purchase over $50 is wild. Side note: you absolutely cannot trust Amazon reviews because sellers routinely give away free product and cash in exchange for five star reviews.

The belief that uttering "communist" is sufficient to win any argument has to be one the funniest recurring archetypes online.

Bark@smorc.bsky.social · 4d ago

Communists for intellectual property rights has to be one of the funniest recurring archetypes online

"That path resolved rng_get() to MicroPython’s software fallback instead of COLDCARD’s hardware RNG implementation." 😂

Watcher Guru@watcher.guru · 6d ago

JUST IN: $40 million worth of Bitcoin stolen after 500 'Coldcard' hardware wallets were hacked.

My computer just woke me up to tell me it's hungry. I'm not kidding 😂 During a long-running task, it noticed the battery was draining, set the volume to 100% using Computer Use, then opened Google Translate and hit "Listen" so I could hear it asking. Pardon my French, but what the fuck? 🤯

Mansplaining School of Prompting defeated. 🤣

mr. TIM@timkellogg.me · 2w ago

them: create a custom harness breakthrough guy: use chatgpt chatgpt.com/share/6a60b2...

Not a good time to be a yak. 🙃

tbabb@tbabb.bsky.social · 2w ago

this yak will be naked

This is a Gen AI free space. I support making Gen AI free.

Playing with tencent's Hy3 (because it's free on openrouter at the minute). Is that a frontier model? I'm guessing it can do 98% of corporate coding jobs... also good enough for Excel-type domain people to use and skip the coder (for appropriate tasks).

End game: hyperscalers get mercy-merged at gunpoint into semi cos.

Dario Perkins @darioperkins.bsky.social · 3w ago

nice summary from Brent Donnelly

Is there a non-infinite-scrolling bluesky client? I think it may help mitigate addictiveness of social media if the "more" operation was multiple choice with no natural default, to increase the cost of the next dopamine hit.

First world problem.

William B. Fuckley@opinionhaver.bsky.social · 3w ago

Increasingly think we need a ‘modernity is actually good’ political tendency. the right is putatively pro-industry and economic growth but thinks science is gay and wants to throw out vaccines for essential oils, the left putatively believes science but also embraces crank degrowth NIMBYism

we are going to make the internet usable again by making it utterly unprofitable. move the web to the art scene

Is github the next stackoverflow?

pi wins again. 🥂 How many billion dollar offers in @mariozechner.at's spam folder?

Sung Kim@sungkim.bsky.social · 4w ago

Benchmarking Coding Agents on Databricks’ Multi-Million Line Codebase They ran a comprehensive evaluation on our tasks, code base, infra. It's been produced by more than 3,000 software engineers, spans 3 hyperscalar clouds and many languages and tasks. www.databricks.com/blog/benchma...

Can we have a code model that can only output code tokens and doesn't "talk"? Like the image models don't spend their time doing "I am going to make the pixel at 302,1207 a bit more yellow"... A classic model could be used post-hoc for documenting what the coder did if needed.

In the maximalist scenario, the hermit who leaves society with weights and self-replicating mechanicals has the same standard of living as the "socialites" who "trade" (what?) with each other.

Ted Underwood@tedunderwood.com · last mo.

Actually sympathetic to Tufekci’s conclusion that we shouldn’t expect jobpocalypse, but people have got to stop grounding that belief in the premise that we have some mysterious power only achievable by protein colloids. Rather: in primate society, lots of roles can only be filled by primates.

Wondering if foundation LLMs being mediocre at creative writing could be due to the corpus being mostly commercial, bureaucratic or scientific, so as to make the small amount of literary material treated as noise (like ungrammatical language is).

the rainbow colors of the pride flag already symbolized the inclusion of everyone, and adding the trans/bipoc/etc colors switched it to enumerating the kinds of people included, which makes it default-exclusive. send tweet

This is how you get id requirements to send bluesky DMs and having to take your shoes off for the airport scanner. Whether or not Sufficiently Advanced AI is harmful to society, there is no credible threat model and matching enforcement mechanism where the enforcement isn’t strictly worse

mr. TIM@timkellogg.me · last mo.

p sure folk here will hate this, but i think that open models above a certain strength (Mythos level??) should indeed be restricted access below that threshold i’m all in on open models, but above leads to some serious social risks idk maybe i’ll think different after significant diffusion

We are but simple people.

The price to pay for persistent context is being permanently silo-ed in one's random life path, living in the prison of truncated possibilities.

Ted Underwood@tedunderwood.com · last mo.

However, sometimes the model may be right to tell you "dummy, we answered this problem in a different thread, and you are now painfully re-inventing a wheel." A couple events like that might make memory worthwhile, no matter how annoying it is.

every single study done on this shows that the actual best behavior is done under consistent pseudonyms and people behave worse under real names it’s counterintuitive but at this point the real names thing is willful denial of evidence

Hannah Posts@hannahposts.bsky.social · 2mo ago

"The problem with the internet is anonymity" ok but people on nextdoor are insane next to their full name and address

Cherishing the window of charming AI mistakes

Maybe for a project like curl, with that size and impact, they should make it permanent? Route reports to big techs with support contracts who do the triage and give the project a clean feed, and incidentally pay for (some of) the maintenance.

Lukasz Olejnik@lukaszolejnik.bsky.social · 2mo ago

curl's lead maintainer announced the project is taking a vacation. Security vulnerability reports will not be processed in July. None. The reason: four months of being buried under AI-generated reports. Anyone who finds a critical bug in July should wait until August. Paid support contracts excepted

More dystopian surveillance state. Science findings should stand on their own and not be evaluated on the purity of the researchers' pedigree.

ORCID@orcid.org · 2mo ago

These are all great reasons for researchers to get started with an ORCID iD. 💫 Others include name disambiguation, name flexibility, and reduced administrative burden. Check out our page for researchers for more: info.orcid.org/researchers/ #ResearchSky