Starts With Frog

28 Aug

If you hate AI with the force of a thousand exploding suns, you'll probably want to block this account. If you work in the field, or are fascinated by AI but deeply troubled by some of the possible long-term consequences to society, then... hi! Nice to meet you. 😊

Wow, so fancy! Claude is writing the manuals for the thing Codex and I built over the past week. So, uh, so I can learn how to use it! 😊

Just for reference, we brought up the hardware for this system yesterday! And we only started the project last week! bsky.app/profile/star...

Starts With Frog@startswithfrog.bsky.social

Saturday morning stuff. Time to test on real hardware!

My ten day old project now has 1200 pages of documentation in three volumes. 🥳🤯

I'm trying to set up my OpenAI "dot" to keep track of my giant pile of projects. Meet "Quill", my new project librarian!

Let's look in the other direction, shall we? (We're actually working on this in our spare time, sort of.)

🎃 Out Of Context International Dubs 🎃

I am a very serious user of artificial intelligence. 😐

Starts With Frog reposted

@outofcontxtdubs.bsky.social

Show: Donald Duck Language: Kurdish Country: Iraq

LOL, good times!

⭐

🐻

Saturday morning stuff. Time to test on real hardware!

View full thread

Running on both boards on the first try! Not a bad start! Good job, Codex.

First mockup of the little web desktop to go with it.

Saturday morning stuff. Time to test on real hardware!

Closeup 👀🔎

ai.nejm.orgComparative Cost of Care Recommended by Large Language Models versus Treating PhysiciansLarge language models (LLMs) are increasingly consulted for clinical decision support, yet the cost of services or treatment they recommend has not been systematically evaluated. We evaluated 24 AI...

And the other board we’re hoping to support. Our original target, actually.

Starts With Frog reposted

eva (^_^)/

@eva.computer

AI doctor: "I see, your concerns are valid. Let's make sure we order a diagnostic to make sure we determine the root cause of what is bothering you." human doctor: "i said there's nothing wrong with you. yeah i'm sure the internet tells you lots of things, get out" *writes drug-seeking in chart*

Scott McGrath@smcgrath.phd

🧪 A new study found care recommended by LLMs costs 3.2x more than plans from treating physicians ($314 vs. $97 per visit at Medicare rates). Across 200 real clinical encounters, AI ordered 5.6x more specialist referrals and nearly 2x as many new medications. #MedSky

Starts With Frog reposted

Made the mistake of asking ChatGPT to go through the unresolved design decisions one at a time. Now I'm stuck typing YES every 30 seconds when I just want to go to bed. 🥱

👵 Wait, you didn't add floating point math yet?!?! 🤖 ... hey, let's add floating point!

ChatGPT has basically finished a thing, but the hardware the thing is intended to run on... hasn't been built yet. And I won't have time to do it until the weekend, probably. Damn thing builds toys faster than I can play with them. 😆/🫠

View full thread

The Challenge: Make an Animated Demo

MOON PARADE! Ok, that's reasonably whimsical. Challenge passed! 🥰

👵 I don't have time to wire everything up right now; can you just add an OLED screen emulator? 🤖 Sure!

"Hey why don't you simulate the colors too; the top 8 rows are yellow I think, and the rest are blue... and mock up some kind of status display."

Back in the olden days it was necessary to explain what "clicking" was... in the manual for a programming language.

The Whitewater Group's "Actor" language from 1987. Came with a special Runtime Edition of Microsoft Windows, version 1.03 or 1.04 or something like that.

Starts With Frog reposted

elynxabeth holmes ⎔

@freyja-lynx.dev

Replying to @freyja-lynx.dev

if anyone in the bay has a trusted mechanic please lmk i’m going to kms

Starts With Frog reposted

Machine keeps OOMing. And Codex is only working on two things: - reinvent software (do a breakthrough) - reinvent hardware (feasibility study) Too much? 🤪

It turns out that Codex’s use of “native” here means not ACTUALLY native, but native to the little virtual machine it’s working with. Apparently it’s written itself a little compiler to run inside! 🤷‍♀️

Testing hardware designs with ngspice. Also it keeps pointing out NOT APPROVED FOR MANUFACTURING which is actually a little creepy. 🤨

Looks like the cat has been doing some AI work of her own.

Starts With Frog reposted

I asked ChatGPT what programming would look like when humans no longer need to look at the code. Long story short, ChatGPT proposed a research project. We built a tiny Lisp implementation in Go, plus an MCP interface for managing multi agent research. First result. Early days!

[Some structures and components it built along the way…]

Starts With Frog reposted

Starts With Frog reposted

Davide Aversa 🎃

@davideaversa.it

I like that LLMs can now experience the quintessential developer experience: going over a codebase, thinking "This is very badly designed!" and discovering it was written by an older version of themselves.

🤔 The really interesting thing here is that “write something related to proof construction and validation” isn’t specifically part of this experiment; this is just what ChatGPT has come up with so far.

👵 Do a breakthrough or something. 🤖 … 🤷‍♀️! Also, yeah, unexpected trees are often a problem. 🌳

Claude wrote something for us tonight but I'm not 100% sure what it even is. "Status: the scratch daemon is running and tools/list shows all 24 tools. I'm now scripting a full three-principal scenario against it to capture real requests and responses for the reference." (It's writing docs. 🤷‍♀️)

Ok I looked at it and it’s actually kind of cool but a lot more elaborate than I expected based on my rather offhand request!

Strange things we say to AI: " Try to save as much cat as possible, but its probably ok to lose part of the frog's butt, nobody will miss it." Trying to get a quick image crop. 🤭

... when your Github account spans two different BlueSky alts.