iiinova — where your digital self keeps evolving

qui sera la demain au Grizzly Fuzz pour la soirée electorale?

000

perhaps because it’s in the early stages and people feel they can still shape the future? Some technologies only become truly harmful when they become pervasive and impossible to escape. It’s not just about individual choices.

000

I hold these two views simultaneously: 1. The combustion engine is useful technology 2. Car-centric cities are dysfunctional and, on balance, harmful to the people who live in them I hold essentially the same view about LLMs

042

Fair enough, I agree with that. And, for the record, I have argued strongly against people who hold the view that LLMs are useless. I guess my perception is being skewed by seeing way more anti AI-criticism in my feed than AI-criticism.

000

you are the one conflating abusive people with, say, Ed Zitron.

000

I'm sorry if i have been overly aggressive in this discussion. I sincerely want to know what kind of policy those people could promote that you'd find a net negative for society as a whole.

100

I also happen to burn hundreds of dollars in tokens every week, so let the "LLMs arent useful" mob come to insult me here. I think I will survive.

000

I'm not a doomer, but I am definitely pessimistic in the short/middle term. I also happen to believe that people who say LLMs aren't useful are waaay, WAAAAY lower on the scale of people causing harm. I don't condone harrasment, obviously.

200

In all honesty, in your heart of hearts, do you believe that anyone who says, “There is no scenario in which LLMs are useful,” has any chance of changing the current trajectory of AI or the AI industry?

100

I think you very greatly over estimate how much of an audience the "LLMs are not useful" camp has in the real world.

100

but ok, I acknowledge that senators discussing policy is not irrelevant. But I very much doubt that, in those discussions, anyone has EVER come close to suggesting that AI may not actually be useful.

000

That discourse is largely irrelevant. The dice are already cast. I’m just annoyed by people getting worked up over bad but relatively harmless takes from people who hold close to zero power

200

no one is going to take your beloved AI models away from you, don’t be silly. the people who hold the power have not only decided that you have the right to use them, they have decided you _will_ use them. No need to be afraid of the anti-datacenter hippies.

000

I’m tired of this kind of both-sidesism. It obscures the enormous asymmetry in power between these groups. yes, there are some anti-AI who dont know what they are talking about. who cares? have you seen the power of the people on the other side?

100

I don’t think those two camps are relevant. They’re the fringes. The real debate is between ordinary people with legitimate concerns (including plenty of AI users) and a small group with immense wealth and political power betting the global economy on AI, consequences and ethical concerns be damned

200

lispers will tell you to use emacs + sly, which I already do, but sometimes you just want to run a repl and evaluate a couple of lines. jfc.

000

It's 2026 and SBCL's repl still doesn't provide a functional line editor.

101

That’s fine if those skills are deprecated by a new technology, but that remains to be seen in the long term.

000

people don’t necessarily feel like their are losing skills (after all they are simply asking ai agents to do things they already know how to do), but if you stop doing something you’ll eventually forget how to do it right and if you stop manipulating the codebase, it will eventually become foreign.

The software industry is currently paying millions of dollars in tokens, degrading its own workforce skills, making the bet those skills will no longer be necessary (or that engs. will keep them up sharp on their own time). That’s a huge bet.

100

wild how we moved from “don’t worry, good engineers will review the code that AI writes anyway” to “code does not matter, it is a compilation target” in a matter of weeks.

000

the thing with microblogging is that it is close to the perfect case for a non-permissioned protocol. For pretty much everything else you need access control and non-public data or people wont use it. So until Atproto Spaces is out, I dont think anything else can come close to bsky

Sergio @sergio.bsky.col.social · 28/09/2026

We’re now in the “writing code by hand considered harmful” stage of AI.

120

Sergio @sergio.bsky.col.social · 27/09/2026

Tascherau devrait être safe quand même. Enfin je voudrais croire. Je m’en vais voter tantôt (vanier-les rivières), ici c’est les conservateurs qui vont probablement gagner.. 🤮

101

Sergio @sergio.bsky.col.social · 24/09/2026

I’m sure there were fringe voices 100% opposed the combustion engine. But I also bet there were people who foresaw the US becoming dysfunctionally car-centric and were dismissed as luddites

000

Sergio @sergio.bsky.col.social · 21/09/2026

Sur ces derniers sujets, je suis convaincu que son autorité scientifique ne lui donne aucun "insight" particulièr d'importance.

Sergio @sergio.bsky.col.social · 21/09/2026

Perso, je pense que Bengio est en même temps un expert et un charlatan: un expert quand il parle de ses propres travaux en ML mais un charlatan des qu'il se met à prophétiser sur l'IA générale ou les risques existentiels

Sergio @sergio.bsky.col.social · 20/09/2026

At the end of the day, the question is: is there still value in humans expressing computation in unambiguous languages?

000

Sergio @sergio.bsky.col.social · 20/09/2026

I just moved the sentiment analysis for my feed generators (slrpr.xyz) from a custom amazon's nova-micro prompt to jev. We'll see. #atproto #jev

slrpr.xyz

SLRPR

012

Sergio @sergio.bsky.col.social · 20/09/2026

Why is SUPER + Q the default shortcut for a terminal in hyprland, that's insane.

001

Sergio @sergio.bsky.col.social · 18/09/2026

Increasingly convinced that napkin notes are now the source and that specs are artifacts akin to lockfiles. Because why not.

000

Sergio @sergio.bsky.col.social · 17/09/2026

personally, I’d like atproto to have protocol-level identity attributes (name, nickname, avatar, email..), independent of the applications. I think openid supported this? oidc certainly does..

Sergio @sergio.bsky.col.social · 15/09/2026

One thing I am convinced of about AI is this: we’re reaping the benefits now, while the real collective costs are being deferred.

010

Sergio @sergio.bsky.col.social · 15/09/2026

my conspiracy theory: AI progress is hitting asymptotic returns and the AI leaders calling for a “slowdown” is just the industry’s way of preemptively justifying why the next models will be only marginally better.

020

Ciega, Sordomuda - Wikipedia

ok, that made me think of Shakira en.wikipedia.org/wiki/Ciega,_...

en.wikipedia.org

000

(fwiw I also believe that the negative reactions to the bluesky38 thing are totally unhinged, of course)

000

*I* did. Consent requires understanding what you're agreeing to. I don't believe a lambda user is being a fool by expecting that his public posts will reach an organic audience instead of being systematically harvested, indexed, profiled and processed at scale by third parties.

200

I miss when a good-looking README or website was actually a decent signal that a project was serious

010

when you said “you consented”

100

why would you expect a random user of bluesky to know about those things.

100

It’s strange that antirez fails to see this. I’m sure he can imagine that (in a similar vein) large numbers of would-be-programmer kids simply aren’t learning to code because they can vibe-code games now. A few years ago they would at least have tried. Most won’t study the generated code.