felipe.meneguzzi.eu
AI Researcher and former AAAI Councilor.
Professor of Computer Science at the University of Aberdeen.
Bridges Professor at PUCRS.
Views expressed here are my own.
https://www.meneguzzi.eu/felipe/
124 posts
112 followers
76 following
Regular Contributor
Active Commenter
comment in response to
post
I just returned from the US for a conference. Seeing videos of this woman at the border together with the security checks in the airport gave me some creepy neo-fascist vibes.
comment in response to
post
Surely you are joking Mr Chollet, how can this titan of innovation, who (re?)invented the Hyperloop (subway), not be ahead of this time?
comment in response to
post
Well, the Roman empire was helped down by a bunch of plagues. Perhaps the American Empire will be helped down by plagues too, seeing that the new government is turning away from science, especially in health.
That was the uplifting thought of the day.
comment in response to
post
I don't think many of the readers got the irony in your message, so I'm just signaling it here...
comment in response to
post
Man in the High Castle...
comment in response to
post
If I may abuse your goodwill, Wayve seems to also have an end-to-end architecture (or so they say), are familiar with the difference between the two approaches?
comment in response to
post
Is the most self indulgent president in American history ready to give the modern equivalent of the malaise speech that helped mess with the presidency of the most personally virtuous president in American history?
comment in response to
post
Yours is much better than the historic one...
comment in response to
post
I think this has to do with religiosity. I see the far left as a manifestation of a non-deistic religion, with the whole faith signalling and whatnot. Americans tend to be more prone to overt demonstrations of faith/religiosity than Europeans. America really likes "Autos da Fé" (check wikipedia).
comment in response to
post
The problem with many hallucinations is that when they happen, even in the most advanced ChatBots, they happen unpredictably, and do so catastrophically. Sure humans can also fail catastrophically, but the solution for a human operator in combat should not be to bullshit your way out of it.
comment in response to
post
But I think there is another kind of deception, which is where makers of LLMs have a vested interest in making their product sound like they are much more capable than they actually are (and the probably know it).
comment in response to
post
Thanks for the vote of confidence here. Also, I think the article on AI deception you posted earlier dovetails with this rather nicely.
comment in response to
post
By that I mean that a lot of the claims of success from LLM companies are based on benchmarks that you can always train your product to ace, even if it the LLM ultimately sucks at anything outside the benchmark. I think was one of the blind spots in the ChatGPT vs DeepSeek surprise.
comment in response to
post
I'm not sure everyone who peddles LLMs as panaceas (or even many decision makers) have such a clear view of their limitations. Even then, while companies work towards reducing detectable hallucinations, there is a lot of debate on the suitability of current assessment methodologies for that.
comment in response to
post
Which is that while LLMs are useful for some tasks where precision is not important, any mission critical application of them (especially where lives are concerned), should be extremely skeptical, because "hallucinations" are simply unavoidable in the stochastic parrots.
comment in response to
post
If you forgive me for harping on this again @shashj.bsky.social, that other post you made on a piece about Wooldridge (and other people at the Turing Institute) "not believing" in AI, this is exactly what the "unbelievers" think.
comment in response to
post
So… give your local university professor a hug (ask permission first) and thank 100+ years of the US finding a really successful and efficient formula for a creating public good that underpins the nation’s economic success.
Here is some more analysis: mark-riedl.medium.com/visualizing-...
comment in response to
post
And I'm not even that worried about my physical safety, I'm just worried a DHS agent might want to confiscate my devices on a power trip, which is something I never expected in my lifetime would happen in the US.
comment in response to
post
I have to attend a conference in the US in about a month and I am very worried about it right now. I have only not cancelled my plans because I feel I'd be letting people down (my co-authors) if I don't go. I'd be very hard pressed to travel to the US optionally right now.
comment in response to
post
While I'm not even close to War Studies, I'm actually surprised that KCL would have such (presumably) fringe views there. I did my PhD in Computing at KCL (and continue to collaborate in that space), and everyone I know in computing is pretty normal.
I take it this is very unconventional in WS?
comment in response to
post
Since you are institutional colleagues, did you ever go chat with him about this?
comment in response to
post
Well, everyone wears headphones these days. I love to work whilst listening to unspeakable shit (lite).
comment in response to
post
Let's bring back child labour too! This way, people can afford to have more kids to supplement the family income!
comment in response to
post
They really want to go back to the early 20th century, when women had a bunch of kids in the expectation that a large percentage of them would die from a combination of infectious diseases and a carelessly protected environment.
comment in response to
post
Each drop of gas in my tank only powers me 0.06% of the way to my destination. Negligible. Therefore, I don't need to pay for any gas.
bsky.app/profile/hype...
comment in response to
post
Seen anything suspicious worth further investigation at your agency?
I take source protection very seriously.
Download Signal (a great encrypted application for the average user), click the little pen and pad, search for my username: JennaMcLaughlin.54.
comment in response to
post
This comes to mind: www.youtube.com/watch?v=f3Zh...