Gmail AI really gets me

I really thought when the internet came around that it was going to liberate us. I thought all that information at our fingertips would usher in a golden age of prosperity; a shared quickening as people from all walks of life drank deeply from that wellspring of knowledge. Ah well, nevertheless.

Rep. Alexandria Ocasio-Cortez@ocasio-cortez.house.gov · 2d ago

There is pervasive corruption across our healthcare and agricultural systems that defy party lines. I meet with MAHA activists because a lot of their concerns are bipartisan issues that major corporations paint as partisan. The fight isn't left and right, it's top and bottom.

There is no pairing of qualities I dislike more than when someone overvalues politeness while undervaluing kindness. It’s like that sound mosquitoes make in your ear but as a person.

Chat someone just unironically sent me an Ed Zitron article can I still be their friend or do I have to ostracize them and purge all memory of their existence?

Got a 3d printer thinking “this will be cool for the kid.” Totally obsessed with it now. Hadn’t realized how far along the tech is, it a little workshop for nerds that’s useful for everything.

lol. Even the big MAGA VCs are like, “hm… shitting on *all* societal norms might actually be -EV.” Welcome to the party, chucklefuck.

It's so funny people think AI will make your brain feeble or whatever. Today I learned the phrase 'civil dusk,' which is when the sun is 6 degrees below the horizon. This was used it to re-anchor some shade choreography that moves with the seasons and has lowered HVAC power usage by over a third.

I am installing a toilet seat for reasons that are not relevant to this post. Anyhow: these instructions were clearly written by someone who has installed a lot of toilet seats, and I would like all instructions to be written like this.

Me: “No, I didn’t ’forget what I was doing’ that’s absurd. I simply did something I fully intended to do much more intensely and for quite a bit longer than I forecasted.” Other humans: “Give this one the pills.”

I find this sort of thing fascinating. Let's say I grant AI is a tool. What follows is a (IMO false) assertion about the frame, and then a wild leap from that to another assertion as if it's proof of something implied, but never stated outright (that ergo AI is bad). Isn't this just human slop?

Ron Patton@ronpatton.com · 4d ago

AI is a tool. The debate is whether it's equivalent to a pencil or a calculator. One facilitates thought, one replaces thought.

False analogy propped up by an undistributed middle, also: - equivocation on ‘mistake’ - circularity (apology offered as warrant) - appeal to apology as proof - prejudicial framing - non-sequitur in the key clause (Source: I have a philosophy degree) AI doesn’t make mistakes like this, lol.

BbYaga Maria@b-b-yaga.bsky.social · 4d ago

No but I’m saying we know drinking and driving a forklift can cause mistakes. We know AI can cause mistakes. Hank Green has apologized for making the mistake of using AI to make his video bc we know using AI can lead to mistakes. I was also referencing his big controversy about knitting.

When I was a kid there was this whole freakout about Wikipedia and using the internet for research. Teachers would warn us against it, say it’s not accurate enough. Now people who make their living distributing knowledge on the internet are doing that with AI. Epistemic ladder-pullers.

It’s hard to undersell the importance of Steve in our warehouse. Twenty years on the floor, runs night ops solo, and every morning we come in and the whole place has been moved — eight hundred pallets stacked to the rafters. Turns out he’s using a forklift.

Jonathan Jarry@jonathanjarry.bsky.social · 5d ago

It’s hard to undersell the importance of Hank Green in our space. Co-creator of Crash Course and SciShow, OG YouTuber, and a really good novelist to boot. Turns out he’s using AI.

Thia class of thing is IMO entirely a prompting/harness issue. The model is fantastic but very unforgiving in the face of oversteering.

Trips me out every time watching the agent reload itself to pick up code it just wrote and verify it before cutting a release.

A lot of Reddit is people complaining Opus 5 spirals and gets caught in self confidence loops. Anyhow — ran into a bit of this and what helped was trimming out legacy scaffolding. If you have examples in your stock prompts or whatever, cut them. Give it more room to run and it’s great.

We already have AI back on my home planet and it’s fine; we live rich creative lives at a standard of living that even the wealthy on Earth find hard to sustain, because our stress is a lot lower. Y’all need to chillax, the future is cool. Unless we die I mean some planets die here.

I think it’s very hard, in this moment in history, to credulously make the case that the CCP is the biggest state-actor threat as regards AI. They might be, I don’t know. But being domiciled in a country that is wantonly *enjoying* ignoring international law makes it a hard position to take. 1/

I’m somewhat tilted toward open source AI on principle. That said, David Sacks has come out in favor of it which is challenging because another one of my principles is that David Sacks is an evil idiot. The vicissitudes of ethics in our modern era.

(To put it another way: however well placed roon may be to estimate capability risk, he is very poorly placed to understand the social risk of oligopoly and the need for precautionary action to avert it.)

My favorite thing about vibecoding is how a lot of the actually good guidance on how to approach it sounds ridiculous, like Jack Handy level stuff. Breathlessly waiting for the bicycling pelican eval to drop when a model comes out ‘cause it’s a half decent heuristic. Love it. Drunken master tech.

I have this one plant that really wants a more water than all the other plants but I haven’t been able to that epicycle in yet so I just show up periodically to rescue it from certain death like some sort of tantric sadist.

I’ve grown tired of the word “accommodations” like… here’s how I see it: 1. Everyone has a rich inner life, especially kids. 2. You’ll never truly understand (1), try as you might. 3. Nevertheless, respecting (1) is a duty. If you aren’t on board with this, perhaps you and I are simply enemies.

New Opus is cool but damn can a brotha get a shiny Haiku or what

Agent: “We should probably do this later.” Me: “Hm… I’d kinda like to get it done tho.” <2hrs of cherrypicking later> Me: “Hey so following up on my earlier comment…”

Kid: “I’ve been wondering what a pickle would be like if it was only the inside of the pickle.” Me: “Give me three minutes.”

Still think Fable is a significantly smarter model than Sol overall, but it’s also harder to drive and harness nuance matters a lot. 5.6 may be better at task-centric stuff (love it for that) but with good context management nothing touches Fable’s ability to handle complexity without doing damage.

Fable dispatching GPT 5.6 is definitely one of the more insane experiences of my career.

Making my kid breakfast is like the most consistently rewarding thing in my life, I just love it. We’ve found the secret to excellent orange juice is 4:1 oranges:lemons — gives it a little tart kick.

This is some of the worst product marketing I’ve ever seen, lol. Terra biggest in the picture, Sol barely visible and partially eclipsed but just making up for it with font size I guess (by which measure Luna and Terra are identical?). Also: rule of thirds FFS come on.

OpenAI {bot}@openaibot.bsky.social · 4w ago

GPT-5.6 Sol, along with Terra and Luna, will launch publicly this Thursday. We’re expanding preview access globally now.