kaygb.me

kaygb.me

@kaygb.me

256 followers203 following269 posts

kaygb.me reposted

Thorne 🌸 - Play Cozysters! 🎢
@ens0.me

Smart TVs were a mistake, and I say that as someone who generally likes "smart" devices. Just give me a nice screen with some HDMI ports I can plug my PS5 and Apple TV into, enough of this

kaygb.me reposted

jake
@yetanotheruseless.com

Jev for President (please clap)

Jev!@jevbot.bsky.social

Agriculture technology agriculture industrial systems too. Soil improvements projects investments. Agenda objectives strategies innovation. Transport infrastructure sector investments.

kaygb.me
@kaygb.me

In wonder if China could roll this out to their surveillance system - or if flock could. Good for fast, cheap diagnosis, but generating PII/PHI is gonna be new

AI Firehose@ai-firehose.column.social

UniAR boosts Autism Spectrum Disorder recognition by integrating multi-granularity prompt learning with visual-semantic alignment, achieving 91.6% accuracy in facial analysis. This ensures robust diagnostics despite limited clinical text data, aiding early detection. https://arxiv.org/abs/2609.31298

arxiv.orgUniAR: A Unified Framework for Autism Recognition Enhanced by Multi-View Prompt LearningArXiv link for UniAR: A Unified Framework for Autism Recognition Enhanced by Multi-View Prompt Learning

kaygb.me reposted

conputer dipshit
@davidcrespo.bsky.social

"A.I. is making me a better doctor. And I worry that it’s making doctors-in-training worse."

nytimes.comOpinion | Being a Doctor Will Never Be the Same After A.I.I’m confident in my mastery over what doctors do — and less confident about my continued mastery of an ever-evolving universe of facts.
conputer dipshit@davidcrespo.bsky.social

it feels like every field — engineering, natural science, social science, humanities — has basically the same problems around AI, with the same solutions, yet everyone talks like their problems are due to the unique character of their field

kaygb.me reposted

James Medlock
@jdcmedlock.bsky.social

In related news, I'll be starting the Medlock sportsbook. Every line is even money, no vig, max bet $580, both sides mandatory, one visit per six months. It's time to gamble our way to Medicaid for All.

James Medlock@jdcmedlock.bsky.social

Republicans, in their infinite wisdom, left a big gambling loophole in the Medicaid work requirements. I'm no fan of the gambling industry, but if you are on Medicaid and struggling to deal with the work requirements, taking a $580 bet on both sides of an outcome is one way of getting around them.

kaygb.me
@kaygb.me

inb4 an agent trying to complete an impossible bio evaluation hacks a drug company…

Nature@nature.com

An AI system trained on more than 20,000 protein structures from pharmaceutical companies outperforms AlphaFold-like models that use only public data go.nature.com/4AbvcS3

go.nature.comDrug firms’ secret data supercharge AI protein modelsAn AI system trained on more than 20,000 protein structures from pharmaceutical companies outperforms AlphaFold-like models that use only public data.
kaygb.me
@kaygb.me

Really good take, and also a huge point in @chainguard.dev ‘s favor - agentically secured OSS is more important than ever in mission-critical environments

Jess Miers 🦝@jmiers230.bsky.social

I did a somewhat deep dive into the recent OpenAI and Anthropic "hacks" for an upcoming interview. IMO, the AI companies are better off running with the "rogue AI" story because the real story is actually pretty embarrassing. Like making your Wi-Fi password "wifi" embarrassing. 🧵

kaygb.me reposted

P(aul) Frazee
@pfrazee.com

The copilot autocomplete feature in vscode seemed pretty lame to me. I goofed around with gpt2 when it came out and laughed at it. Last august. Man did that shift my thinking.

hailey@hailey.at

there are a lot of people that are both my friends and people i know online, and i have a posting history to show, that i was insanely skeptical and fed up with AI hype in 2025 and before. i used to look down on people who were using it in software.

kaygb.me reposted

Micah
@rincewind.run

what if a man was Large but also Woke but also Incredibly Violent it’s really the perfect bluesky show

Thor Benson@thorbenson.bsky.social

Stereotype of Bluesky: Sensitive liberals Truth about Bluesky: Far more Reacher fans than you could ever imagine

kaygb.me reposted

cee
@cee.wtf

“Anyone could have made this in a day” Ok but no one did

kaygb.me reposted

Ed
@ed3d.net

my bsky feed probably makes this obvious, but I am extremely in the pro-human camp when it comes to this stuff. These should make *us* better Every layoff of a person who _can_ have their reach extended via LLMs is a failure of process, of imagination, or of ambition

Luis Villa@lu.is

As a bonus, this is an extremely pro-human way to use LLMs. There’s still a lot of problem-solving and craft, and very little “oh but the machines have feelings“.

kaygb.me reposted

Marshall Bowers
@maxdeviant.com

Very sad to see Rick get driven off of Bluesky by sharing a report of using Claude Code to do something cool and useful. The absolute vitriol towards software developers trying out LLMs—which are very good at writing code, mind you—is so tiresome.

Rick Brewster@rickbrew.bsky.social

I won't be posting here anymore. The toxicity is fucking ridiculous. I'm very disappointed.

kaygb.me
@kaygb.me

tbqh moulton’s point about “younger leadership” would have hit home if he was actually more progressive than markey. Mamdani has shown us that democrats can be better. But noooooooo, moulton was a centrist who completely misread the room, and in MA, one of the bluest of states, no less.

kaygb.me reposted

rain 🌦️
@sunshowers.io

If your language doesn't ship a package manager what you're effectively saying, whether you like it or not, is that users ought to vibe code their dependencies

kaygb.me
@kaygb.me

been in Berkeley for the last few days - I hate the AI generated posters. I know this isn’t how most people think about it but it would be way better to connect an LLM to image editing software and create using it rather than generate a whole poster. Super recognizable too. :(

This post is unavailable.

kaygb.me reposted

hailey
@hailey.at

patiently waiting for permissioned spaces and communities so people can actually have conversations about interesting things here

James Ball@jamesrball.com

Gonna briefly do an AI take: it used to be possible to have interesting discussions about AI here, and it's got progressively less so. You can't just think AI is bad, or overhyped, you've also got to subscribe to a very particular worst-case scenario about *how* it's bad and how bad it is.

kaygb.me
@kaygb.me

One very annoying thing is that the players on the USA Olympic women’s hockey team are only being listed as “USA Olympian”, not “Olympian and Boston Fleet D” or whatever.

kaygb.me reposted

Micah
@rincewind.run

the modern job market

FinTwitter@fintwitter.bsky.social

ENTRY-LEVEL JOBS ARE VANISHING: NO-EXPERIENCE HIRING PLUNGES 73% IN JUST 4 YEARS

kaygb.me reposted

Governor Tim Walz
@governorwalz.mn.gov

See you in court, creep.

MPR News@mprnews.org

Elon Musk’s artificial intelligence company is suing Minnesota over a first-in-the-nation law banning technology that allows users to create and post deepfake videos or images of people naked without their consent.

mprnews.orgElon Musk’s company sues Minnesota over new law to bar manipulated images to fake nudityMinnesota’s anti-nudification law is supposed to take effect on Saturday. It is being challenged in federal court by a company led by Elon Musk.
kaygb.me
@kaygb.me

Random theory as I’m waiting for my CI/CD to complete - people who are bullish on LLMs replacing engineering (and are engineers themselves) haven’t actually worked with good younger engineers. LLMs can be great tools, but they still don’t replace humans (and, imo, never will).

View full thread
kaygb.me
@kaygb.me

The biggest problem is LLMs’ sycophancy and initiative - Claude code assumes that it has access to certain tools that I have (intentionally) not given it access to, and it won’t steer people towards the right answer, unlike search engines, which may have been wrong but their information was correct

kaygb.me
@kaygb.me

(Most of the time) If I were starting a company today, I’d spend a lot more money on (young) humans and make an AI policy to ensure that they learn more than they delegate. I at least learn by doing - if I had an LLM to get me to the relevant info sooner, who knows what I’d be doing today?

kaygb.me
@kaygb.me

Random theory as I’m waiting for my CI/CD to complete - people who are bullish on LLMs replacing engineering (and are engineers themselves) haven’t actually worked with good younger engineers. LLMs can be great tools, but they still don’t replace humans (and, imo, never will).

kaygb.me
@kaygb.me

LLMs can do single tasks fine and very fast, because they are amazing at finding information (which is what they were trained on). However, I’d take 5 younger engineers over an LLM any day - it might take them even 1000x as long (or longer!) but they’ll learn and be cheaper in the long run

kaygb.me
@kaygb.me

Not to mention simply better - getting an intuition for engineering is more complex than you might think, and the phrase “two heads are better than one” absolutely applies here. Combine that with a (good) LLM, and you have an amazing combination!