Ásgeir Berg
@asgeirberg.bsky.social
Reposted by Ásgeir Berg
Despite writing about the details of LLMs, and how they cannot be trusted, I recently got bamboozled by some elaborate hallucinations. 🧵
October 30, 2025 at 11:47 AM
Despite writing about the details of LLMs, and how they cannot be trusted, I recently got bamboozled by some elaborate hallucinations. 🧵
Reposted by Ásgeir Berg
New paper! 🥳
"Trust in Mathematics", coauthored with Silvia De Toffoli is out now in Philosophia Mathematica.
A thread on what it is about. Please share with anyone who may be interested!
academic.oup.com/philmat/adva...
"Trust in Mathematics", coauthored with Silvia De Toffoli is out now in Philosophia Mathematica.
A thread on what it is about. Please share with anyone who may be interested!
academic.oup.com/philmat/adva...
Trust in Mathematics
Abstract. In this paper we develop a systematic account of trust in mathematics based on Katherine Hawley’s commitment account of trust. We focus on two re
academic.oup.com
October 29, 2025 at 12:50 PM
New paper! 🥳
"Trust in Mathematics", coauthored with Silvia De Toffoli is out now in Philosophia Mathematica.
A thread on what it is about. Please share with anyone who may be interested!
academic.oup.com/philmat/adva...
"Trust in Mathematics", coauthored with Silvia De Toffoli is out now in Philosophia Mathematica.
A thread on what it is about. Please share with anyone who may be interested!
academic.oup.com/philmat/adva...
Reposted by Ásgeir Berg
Being an externalist now is so freeing. When someone points out that people with evidence identical to mine have been wrong in the past, I can just say "The difference between them and me is that I'm right"
October 29, 2025 at 6:31 PM
Being an externalist now is so freeing. When someone points out that people with evidence identical to mine have been wrong in the past, I can just say "The difference between them and me is that I'm right"
My paper defending nominalism about mathematical objects is out in Synthese: link.springer.com/article/10.1...
How the laws of logic lie about mathematical objects - Synthese
In ontological debates on the existence of mathematical objects, it has long been taken for granted that if we take our mathematical discourse at face value, it follows from the fact that our true mathematical statements refer to mathematical objects that mathematical objects exists; reference in true statements entails existence. In this paper, I argue that there are positions available in the philosophy of logic that allow us to dislodge this assumption, allowing for a nominalist position according to which statements such as ‘there are four prime numbers between 1 and 10’ are true and genuinely refer to numbers, without a corresponding statement asserting the existence of numbers following from it. Consequently, there is an overlooked nominalist position according to which objects just are what singular terms refer to and that reference is successful when those singular terms figure in true statements, with corresponding existence statements being literally false. I argue that this view is not only coherent, but that it does not entail that there are objects that do not exist.
link.springer.com
October 7, 2025 at 8:18 PM
My paper defending nominalism about mathematical objects is out in Synthese: link.springer.com/article/10.1...
I had some unbaked (and perhaps incoherent) thoughts on the Liar Paradox:
asgeirberg.substack.com/p/some-unbak...
asgeirberg.substack.com/p/some-unbak...
October 6, 2025 at 9:20 PM
I had some unbaked (and perhaps incoherent) thoughts on the Liar Paradox:
asgeirberg.substack.com/p/some-unbak...
asgeirberg.substack.com/p/some-unbak...
People don't talk enough about how Google forces a bullshit AI summary on its users with every search. Anything anyone searches for comes with a heap of falsehood.
Given Google's prominence in society, this should be considered an outrage by the general public.
Given Google's prominence in society, this should be considered an outrage by the general public.
October 3, 2025 at 1:26 PM
People don't talk enough about how Google forces a bullshit AI summary on its users with every search. Anything anyone searches for comes with a heap of falsehood.
Given Google's prominence in society, this should be considered an outrage by the general public.
Given Google's prominence in society, this should be considered an outrage by the general public.
Reposted by Ásgeir Berg
Amateur view. AI bubble shows that there is too much money chasing too few opportunities for genuine innovation. The solution is to take a medium/long view and put much more funding into basic science in universities rather than cutting as is happening now. (But I would say that wouldn’t I?)
September 30, 2025 at 8:52 AM
Amateur view. AI bubble shows that there is too much money chasing too few opportunities for genuine innovation. The solution is to take a medium/long view and put much more funding into basic science in universities rather than cutting as is happening now. (But I would say that wouldn’t I?)
Mine was also extremely anti-climactic. I finished it in the beginning of 2020 and then the pandemic rolled around.
The defence was on Zoom and meeting anyone for a celebration was illegal.
The defence was on Zoom and meeting anyone for a celebration was illegal.
Oh, I meant to mark that it's 15 years since my dissertation defence. Possibly the most weirdly anti-climactic event of my life - I was so nervous I didn't sleep at all the night before, and then I got there and no-one on my committee seemed all that inclined to ask me hard questions.
September 17, 2025 at 7:13 AM
Mine was also extremely anti-climactic. I finished it in the beginning of 2020 and then the pandemic rolled around.
The defence was on Zoom and meeting anyone for a celebration was illegal.
The defence was on Zoom and meeting anyone for a celebration was illegal.
"It just predicts the next token based on data in the training set" is a poor abstraction of LLMs because it makes poor predictions
For example, we might predict that it's very difficult to have an LLM emit the phrase "the quick brown fox jumps over the lazy cvnpmnzq", but it's trivial
For example, we might predict that it's very difficult to have an LLM emit the phrase "the quick brown fox jumps over the lazy cvnpmnzq", but it's trivial
September 5, 2025 at 12:00 PM
Can we lower the limit to just "invaded"?
I think if Britain colonised your country of origin you should get free BBC iplayer.
August 29, 2025 at 7:47 PM
Can we lower the limit to just "invaded"?
Two Cellinis but not a single Wittgenstein.
going to go out on a limb and imagine this guy's not actually very good at conversations
August 17, 2025 at 1:42 PM
Two Cellinis but not a single Wittgenstein.
Machine men with machine minds and machine hearts.
The first time Charlie Chaplin used his voice in a film, he wielded it to forcefully condemn the rising forces of fascism around the globe. Will we listen to his words today?
August 14, 2025 at 5:07 PM
Machine men with machine minds and machine hearts.
I really like this post. I'm a quintessential „particularist“—I try to never bite any bullets. Until now, this was an unexamined attitude, rather than a conscious methodological choice.
Biting Bullets
I’ve agreed to review Timothy Williamson’s Overfitting and Heuristics in Philosophy for Notre Dame Philosophical Reviews. This post is not that review. But I will use it as a chance to talk about some...
grecowansley.substack.com
August 8, 2025 at 4:45 PM
I really like this post. I'm a quintessential „particularist“—I try to never bite any bullets. Until now, this was an unexamined attitude, rather than a conscious methodological choice.
What about this: If to infer q from p, one must recognise that p follows from q and for that reason come to believe that q, then that seems a lot like an application of a meta-rule, if "p follows from q and p, believe q".
Don't I then need to do the same again, apply the meta-rule to my meta-rule?
Don't I then need to do the same again, apply the meta-rule to my meta-rule?
I think so, but I now think that my objection isn't quite as straightforward as I thought.
July 30, 2025 at 4:09 PM
What about this: If to infer q from p, one must recognise that p follows from q and for that reason come to believe that q, then that seems a lot like an application of a meta-rule, if "p follows from q and p, believe q".
Don't I then need to do the same again, apply the meta-rule to my meta-rule?
Don't I then need to do the same again, apply the meta-rule to my meta-rule?
I've never quite got Boghossian's taking condition on inference, which says that S can only infer q from p if S takes p to support the truth of q.
Surely it is the other way around, that S can only take p to support the truth of q if S has inferred that q from p?
Surely it is the other way around, that S can only take p to support the truth of q if S has inferred that q from p?
July 30, 2025 at 12:38 PM
I've never quite got Boghossian's taking condition on inference, which says that S can only infer q from p if S takes p to support the truth of q.
Surely it is the other way around, that S can only take p to support the truth of q if S has inferred that q from p?
Surely it is the other way around, that S can only take p to support the truth of q if S has inferred that q from p?
is a skilful player (of Settlers of Catan). A weak player is someone who has the desire to trade sheep around, and perhaps does, but does not yet possess the ore needed to build a city. Have fun, colleagues, but play well.
July 29, 2025 at 7:50 PM
Here is Gary Marcus, pointing out that some of the answers that OpenAI's model gave in the IMO are less than impressive.
But when it comes to LLMs, I'm always more impressed by the mistakes than the successes. The fact that it gave an answer like this, means that it came up with it itself.
But when it comes to LLMs, I'm always more impressed by the mistakes than the successes. The fact that it gave an answer like this, means that it came up with it itself.
July 23, 2025 at 8:51 AM
Here is Gary Marcus, pointing out that some of the answers that OpenAI's model gave in the IMO are less than impressive.
But when it comes to LLMs, I'm always more impressed by the mistakes than the successes. The fact that it gave an answer like this, means that it came up with it itself.
But when it comes to LLMs, I'm always more impressed by the mistakes than the successes. The fact that it gave an answer like this, means that it came up with it itself.
I have a new paper forthcoming in Synthese, "How the Laws of Logic Lie About Mathematical Objects": www.asgeirberg.org/papers/how-t...
I argue that if one adopts the view that there are no logical laws, then there is a coherent nominalist position available that takes our discourse at face value.
I argue that if one adopts the view that there are no logical laws, then there is a coherent nominalist position available that takes our discourse at face value.
www.asgeirberg.org
July 18, 2025 at 12:34 PM
I have a new paper forthcoming in Synthese, "How the Laws of Logic Lie About Mathematical Objects": www.asgeirberg.org/papers/how-t...
I argue that if one adopts the view that there are no logical laws, then there is a coherent nominalist position available that takes our discourse at face value.
I argue that if one adopts the view that there are no logical laws, then there is a coherent nominalist position available that takes our discourse at face value.
Reposted by Ásgeir Berg
Publish or perish: the untold story of Socrates
May 5, 2025 at 9:56 AM
Publish or perish: the untold story of Socrates
Me, when I see Trump's make-up:
April 10, 2025 at 6:30 PM
Me, when I see Trump's make-up:
"Exceptional will" meaning "got a large inheritance".
The worst catastrophes of history are caused by men who are malevolent, ignorant, and have exceptional wills.
April 8, 2025 at 11:40 AM
"Exceptional will" meaning "got a large inheritance".