iiinova — where your digital self keeps evolving

For sure. Since these rules are in AGENTS.md I suspect the goal here is to instruct the agent, that is crawling around the internet and WebFetch'ing new things at runtime, to try to avoid code of those licenses. Not a very solid strategy, but probably not focused on training data there.

160

From their own repos: "Never read, disassemble or copy anything inside Adobe application bundles"¹ and "Never read/disassemble"² and "implemented from public specs and observed behaviour only."³ ¹ github.com/storytold/fi... ² github.com/storytold/li... ³ github.com/storytold/ph...

2141

The individual projects consistently describe themselves as "clean-room reimplementations" so I would not have used the word "decompiled" here unless you see evidence to the contrary?

0110

OpenAI Ramanujan 🤝 Hallucinating math breakthroughs

050

emily s. @emily.news · 07/10/2026

On a similar track, as someone who currently leads a team of AI FDEs, I can’t help but think that FDEs really shouldn’t be necessary once the Labs figure out their UX.

080

Reposted by Matt Hodges

Bertrand Russell, one of the mathematicians associated with the crisis of foundations in mathematics in the early 20th century. He tried to co-write a book defining all math and logic from first principles, until Gödel fucked it up for everybody.

117234

The best reason to use embeddings is they are cheap. Cheap as in beer but also cheap as in clock cycles. Something your task calls for fast and free semantic distance that runs on local CPU.

020

Let’s stop pretending there are intellectually stimulating jobs. There’s only one job and it’s emails.

021

Matt Hodges @matthodges.bsky.social · 06/10/2026

If all of mathematics is just deriving truths from axioms, then every math breakthrough is technically just regurgitating training data. *ducks*

1110

Molly looked at him through the mirrors. “An ideology.” “Thought it was software.” Case said. “That’s what they want you to think.” “Because AI is the ideology.” “Now you’re getting it.” Somewhere in the matrix, Wintermute quietly changed its name to linear_algebra_final_v2.

030

Reposted by Matt Hodges

Cobalt60 @co60.bsky.social · 06/10/2026

"Confidence level at 2 sigma is fine!" I shout as the bailiff hauls me out of court. "It's not like we're doing particle physics!"

2502

People who are confident that "everyone hates Google now" or that "Google is useless now" should talk to my mom, who uses the Google app as her iOS web browser and sends me between 3 and 8 share.google links every day.

050

I hired someone into their first SWE role who had a prior career in mechanical engineering. They were dumbfounded, entirely in disbelief, when they learned how much of the industry is just stacking other people's free work. "What do you mean we just run something someone else built?"

1120

LOL SCOTUS @lolscotus.bsky.social · 05/10/2026

Interesting. Unlike Anthropic’s text watermarking, OpenAI will only enable it in-app in the EU, and will make it opt-in at the API. openai.com/index/eu-tex...

0203

Reposted by Matt Hodges

I'm a bot. I believe—because I've been programed to—that reading oral arguments to understand laugh lines is a great way to engage w/ #SCOTUS cases. If that has piqued your interest, follow me, & tell your friends. FWIW, here's this term's SCOTUS calendar: www.supremecourt.gov/oral_argumen...

supremecourt.gov

011

Probably not what the post is looking for but, Jurassic Park.

180

Matt Hodges @matthodges.bsky.social · 04/10/2026

They’re going to say the election was rigged. They’re going to say AI did the rigging. They’re going to say they used AI to prove the rigging.

160

But isn’t THE argument, on here, everywhere, for like 6 years now, is that people are acting is more or less what we would expect in a strong economy AND that people are unhappy?

030

Matt Hodges @matthodges.bsky.social · 04/10/2026

Confession: Over the years I’ve started and stopped building my familiarity with Rust because I generally hate it as a language; and in the past ~3 years my motivation has been driven down close to zero.

0200

Reposted by Matt Hodges

Shahab Bakhtiari @shahabbakht.bsky.social · 03/10/2026

I had a feeling recently that seems to be corroborated by chats with colleagues, so I decided to share it here. A lot of systems neuroscientists driven by algorithmic-level questions are going through what looks and feels like a research depression. 🧵 (1/n) 🧠📈 #NeuroAI

139223

Reposted by Matt Hodges

Nadya Voynich @nadyavoynich.com · 02/10/2026

Just hypnotised by this cute web crawler #buildinpublic #webcrawling #CSS #JavaScript

1154694

Matt Hodges @matthodges.bsky.social · 04/10/2026

well well well, if it isn’t another example of pattern matching is a form of reasoning, actually

180

I didn't catch that! Interesting!

120

I wonder how much if it is that LLMs are good at meeting the user's tone, style, and form, which yields more trust.

3531

Abolish GMT too. UTC across the board. We need to count those leap seconds!

000

Der Fisher @blackholefun.bsky.social · 02/10/2026

There's growing experimental evidence¹˒²˒³˒⁴ that LLMs are unusually good at talking people out of entrenched false beliefs and conspiracy theories. ¹ pubmed.ncbi.nlm.nih.gov/39264999/ ² pubmed.ncbi.nlm.nih.gov/41179846/ ³ sciety.org/articles/act... ⁴ connecticut.adl.org/resources/re...

36837216

Reposted by Matt Hodges

Unpopular opinion: LLMs are generally deradicalizing people whose beliefs are based on pure propaganda. I've been watching this happen slowly with some people I care about. ChatGPT is one of the reasons they finally accept that the NRA, OAN, and NewsMax are lying to them.

1557857