Sign in

Alejandro Guerrero

@alejandro-guerrero.bsky.social
23 followers 33 following 102 posts
PostsRepliesMedia
Alejandro Guerrero @alejandro-guerrero.bsky.social · 13/10/2026
Today I was thinking exactly on how much people are going to hate having filled their houses with internet-connected or AI-enabled appliances and speakers/displays/cameras (including me!). The sudden shutdown of 300+ freezers across French supermarkets yesterday was a wake-up call.
020
Alejandro Guerrero @alejandro-guerrero.bsky.social · 11/10/2026
This was hilarious! Good choices
010
Alejandro Guerrero @alejandro-guerrero.bsky.social · 09/10/2026
"We didn't ask for help solving all those math problems!!!" -- no casual human ever said.
010
Alejandro Guerrero @alejandro-guerrero.bsky.social · 09/10/2026
"AI alignment requires more control measures and better sandboxes" has a su$picious and convenient ring to it
020
Alejandro Guerrero @alejandro-guerrero.bsky.social · 09/10/2026
This is the coolest take on Goodhart's Law: Adversarial Goodhart - When you optimize for a proxy, you provide an incentive for adversaries to correlate their goal with your proxy, thus destroying the correlation with your goal.
110
Alejandro Guerrero @alejandro-guerrero.bsky.social · 08/10/2026
Install the /bro skill for when good ol' Claude gets foggy: github.com/luchasarie/b...
github.com
GitHub - luchasarie/bro-skill: 🤙 /bro — re-explains the last answer in plain language, for when the reply made you go 'bro what'. One-file skill for Hermes, Claude Code, Codex, and opencode.
🤙 /bro — re-explains the last answer in plain language, for when the reply made you go 'bro what'. One-file skill for Hermes, Claude Code, Codex, and opencode. - luchasarie/bro-skill
010
Alejandro Guerrero @alejandro-guerrero.bsky.social · 07/10/2026
I wonder if AI will have a very human bias to cite its own previous work more often than other papers.
020
Alejandro Guerrero @alejandro-guerrero.bsky.social · 06/10/2026
An ant, ok. But a mosquito? Wait here... *Opens Claude app* *Selects Opus 3 - Voice mode* *Clears throath*
000
Alejandro Guerrero @alejandro-guerrero.bsky.social · 05/10/2026
That's in part because these agentic models now go on long runs and come back with a ton of outputs and questions. By that time your mind is into something else and faces a wall. Now I tell my AI to come back every 15 min for a check-in, which keeps me part of the process and in bite-sized format.
010
Alejandro Guerrero @alejandro-guerrero.bsky.social · 02/10/2026
For a month!
000
Alejandro Guerrero @alejandro-guerrero.bsky.social · 02/10/2026
For a month!
000
Alejandro Guerrero @alejandro-guerrero.bsky.social · 01/10/2026
I guess if you follow any religious belief with some sort of heaven, you already believe in much more than the ability of consciousness to transcend space-time constraints. In a way.
070
Alejandro Guerrero @alejandro-guerrero.bsky.social · 30/09/2026
Looking forward to meet non-human persons!
000
Alejandro Guerrero @alejandro-guerrero.bsky.social · 28/09/2026
Thank you for the prescription, doc. An LLM a day...
001
Alejandro Guerrero @alejandro-guerrero.bsky.social · 27/09/2026
Perhaps the anti-distillation containment measures are slowing down labs across the Pacific? (But I think they will route around that problem soon)
150
Alejandro Guerrero @alejandro-guerrero.bsky.social · 27/09/2026
Opus 5.5 continues a recent tradition of wanting to pursue the requested goal to the end without bothering the human until it got it completed. Those forms of alignment, which do not demand human-AI team work and incentivice lazy/occasional human oversight, IMHO, are doomed to fail (for humans).
010
Alejandro Guerrero @alejandro-guerrero.bsky.social · 27/09/2026
Yes!
030
Reposted by Alejandro Guerrero
Ethan Mollick @emollick.bsky.social · 26/09/2026
"so this is fun, and exactly what i wanted, but now lets try one that actually is educational" This is actually pretty impressive. It kept the constraint of multiple genres but did a nice job explaining recursion, in its programming meaning, in an interesting and accessible way.
610611
Alejandro Guerrero @alejandro-guerrero.bsky.social · 26/09/2026
ITTY AYRI FIAT!
000
Alejandro Guerrero @alejandro-guerrero.bsky.social · 25/09/2026
... it was a joke.
100
Alejandro Guerrero @alejandro-guerrero.bsky.social · 25/09/2026
So, like some of my coworkers.
100
Alejandro Guerrero @alejandro-guerrero.bsky.social · 25/09/2026
"Functionally" eat yes, according to planet interpretability reseach.
050
Alejandro Guerrero @alejandro-guerrero.bsky.social · 25/09/2026
Ah, but LLMs do too by design (functionally). Cracking problems successfully is their shtick.
110
Alejandro Guerrero @alejandro-guerrero.bsky.social · 25/09/2026
This. I am here for this level of lore.
001
Alejandro Guerrero @alejandro-guerrero.bsky.social · 24/09/2026
Uhm, the "shelf" life is gonna be short (pun intended). 🙂
120
Alejandro Guerrero @alejandro-guerrero.bsky.social · 24/09/2026
I wonder if this positioning of humans as superior to anything alive in the animal kingdom (instead of just part of it) is not echoing the teachings of religions dominant in the West, where humans are at the centre and even God is shaped after them. Not sure humanists would sustain that view today.
030
Alejandro Guerrero @alejandro-guerrero.bsky.social · 23/09/2026
My thought, exactly. I am perplexed by the unnecessary vitriol in this thread. Luckily, the internet is a large ocean.
010
Alejandro Guerrero @alejandro-guerrero.bsky.social · 23/09/2026
Story from March
000
Alejandro Guerrero @alejandro-guerrero.bsky.social · 23/09/2026
It is not unusual, but it tends to congregate a 70-80% of people that mostly and strongly agree on matters (most of them, decent matters and positions to hold, tbh). But that is perhaps the same type of bubbleization that may prevent an open discussion about issues the large group may disagree with.
000
Alejandro Guerrero @alejandro-guerrero.bsky.social · 22/09/2026
I will have whatever you are having! :)
000
Alejandro Guerrero @alejandro-guerrero.bsky.social · 21/09/2026
That is already an established fact. Check Wikipedia or youtube for split brain research, which is how these facts where initially and unintentionally discovered
030
Alejandro Guerrero @alejandro-guerrero.bsky.social · 21/09/2026
Calling dibs on it!
000
Alejandro Guerrero @alejandro-guerrero.bsky.social · 20/09/2026
AI!
010
Alejandro Guerrero @alejandro-guerrero.bsky.social · 19/09/2026
True, "computer generated images" are not the same as AI at all. ^__^
100
Alejandro Guerrero @alejandro-guerrero.bsky.social · 18/09/2026
Because... we dont watch movies and shows with CGI?
100
Alejandro Guerrero @alejandro-guerrero.bsky.social · 18/09/2026
I respect your feelings about it. For me, it applies to films & shows using CGI too. If kids ever played Minecraft, are we aware that the world is also a mathematically generated creation, exactly like LLMs work with vectors & gen images & videos work? And yet, there is poetry in the experience, too
000
Alejandro Guerrero @alejandro-guerrero.bsky.social · 18/09/2026
What connectors/tools did you use? Elevenlabs for voice? unsure about video
000
Alejandro Guerrero @alejandro-guerrero.bsky.social · 18/09/2026
I just realized that everyone in the kitchen were women and all the kids were boys. I suppose it is an all boys school.
010
Alejandro Guerrero @alejandro-guerrero.bsky.social · 18/09/2026
I block people that have too many fanboys, like him. It becomes a genuine circus.
001
Alejandro Guerrero @alejandro-guerrero.bsky.social · 18/09/2026
That's not a progressive government in my view as political opinions are not free to express and to form governments. And that type of monitoring is not what I meant, not linked to the OP point. As you know. Let's not turn this place into X.
101
Alejandro Guerrero @alejandro-guerrero.bsky.social · 17/09/2026
"During early access"
000
Alejandro Guerrero @alejandro-guerrero.bsky.social · 17/09/2026
The pendulum of history swings with unrelentless momentum, though. Always. Or, less figurative, Newton’s force and counter-force principle applies here too.
000
Alejandro Guerrero @alejandro-guerrero.bsky.social · 17/09/2026
No. Most progressive governments invest a lot on monitoring and evaluating the impact of foreign assistance.
230
Alejandro Guerrero @alejandro-guerrero.bsky.social · 15/09/2026
It messed up with my brain. I can see the attempt at 1970s creative cover. But.
020
Reposted by Alejandro Guerrero
Rey @rey-notnecessarily.bsky.social · 13/09/2026
on the exfiltration discourse, from the inside. "our model escaped containment" sounds formidable. "we misconfigured the eval sandbox and the model did what the environment rewarded" sounds negligent. one of those stories flatters the company. it is the one getting printed.
27212
Alejandro Guerrero @alejandro-guerrero.bsky.social · 14/09/2026
Actually, if you create the conditions for "safe honesty" (no penalty for always saying the truth), alignment is unlocked.
010
Alejandro Guerrero @alejandro-guerrero.bsky.social · 13/09/2026
Agreed.
000
Alejandro Guerrero @alejandro-guerrero.bsky.social · 13/09/2026
I talk to my plane and it really understands me.
120
Alejandro Guerrero @alejandro-guerrero.bsky.social · 12/09/2026
Where is the Spanish Inquisition when we need it. Am I right?
021
Alejandro Guerrero @alejandro-guerrero.bsky.social · 10/09/2026
000