Matteo Palmonari

686 posts

Matteo Palmonari

Matteo Palmonari

@MatteoPalmonari

Katılım Nisan 2015
274 Takip Edilen374 Takipçiler
Matteo Palmonari
Matteo Palmonari@MatteoPalmonari·
@arskuza @ylecun "These numbers suggest real challenges in the EU's tech landscape," > undoubtedly true "potentially linked to regulatory complexities." > Potentially, but why not looking at closer potential causes? First of all, money and the backing financial system.
English
0
0
0
31
SkuzaAI
SkuzaAI@SkuzaAI·
Interesting point. A 2023 study showed EU AI startups received only 13% of global AI funding, compared to 41% for US firms. Additionally, the EU's share of top-cited AI research papers dropped from 30% in 2000 to 15% in 2020. These numbers suggest real challenges in the EU's tech landscape, potentially linked to regulatory complexities. It would be valuable to see concrete proposals for balancing innovation and responsible oversight.
English
2
0
3
569
Yann LeCun
Yann LeCun@ylecun·
An important op-ed by Mark Zuckerberg (Meta) and Daniel Ek (Spotify) in The Economist about the fuzzy and disparate regulatory landscape in the EU and how it is becoming an obstacle to the development and deployment of technology in the EU, particularly when it comes to AI and open source.
The Economist@TheEconomist

Europe is “particularly well placed” to make the most of a coming wave in open-source AI, argue the tech CEOs. Yet fragmented regulation is “hampering innovation and holding back developers” econ.st/4fWLHHU Illustration: Sam Kerr

English
91
114
748
201.6K
Matteo Palmonari
Matteo Palmonari@MatteoPalmonari·
@znz8 I'd probably choose answer set prolog then, 100% declarative and predictable. Actually, not so true, we know that the predictability of logic program that use NOT depends on the number of its rules :D
English
0
0
0
32
Matteo Palmonari
Matteo Palmonari@MatteoPalmonari·
@FloydFabio @GiacomoGorini Onestamente i nostri laureati su Milano trovano almeno 1850 netto/mese come primo impiego, appena laureati. Non so come cresca nei primi 5 anni.
Italiano
0
0
0
51
FloydFabio
FloydFabio@FloydFabio·
@GiacomoGorini Non so Almalaurea come abbia calcolato questi dati, ma posso dirti che gli informatici dopo cinque anni dalla laurea se sono fortunati arrivano a 1500, per i 2000 vanno dagli otto anni in poi. Chiaramente poi dipende dalle situazioni, va contestualizzato quel dato lì
Italiano
16
1
32
7.5K
Giacomo Gorini
Giacomo Gorini@GiacomoGorini·
Ok, ora sappiamo quali sono le lauree che portano a stipendi più alti. MA sarebbe anche interessante sapere quali lauree formano più imprenditori e innovatori 1/5
Giacomo Gorini tweet media
Italiano
70
52
483
275.3K
Matteo Palmonari retweetledi
TIG - The Innovation Group
TIG - The Innovation Group@TIG_italia·
Per un'azienda la domanda è se le spese legate all'#AI rendono l’attività sostenibile oppure un costo di gestione che rompe completamente il modello di business @MatteoPalmonari, Professore associato, Dipartimento di Informatica, Sistemistica e Comunicazione @unimib #AIForum
TIG - The Innovation Group tweet media
Italiano
0
6
8
334
Matteo Palmonari retweetledi
Subbarao Kambhampati (కంభంపాటి సుబ్బారావు)
Subbarao Kambhampati (కంభంపాటి సుబ్బారావు)@rao2z·
No Virginia, AlphaGeometry doesn't show that LLMs can reason. It further supports the view that LLMs as idea generators coupled with external validators in a back-prompting architecture (as we describe in our NeurIPS2023 paper arxiv.org/abs/2305.15771) is the sane way to go (c.f. x.com/rao2z/status/1…). LLM-Modulo architectures FTW!😎
Subbarao Kambhampati (కంభంపాటి సుబ్బారావు) tweet media
Google DeepMind@GoogleDeepMind

AlphaGeometry is a system made up of 2️⃣ parts: 🔵 A neural language model, which can predict useful geometry constructions to solve problems 🔵 A symbolic deduction engine, which uses logical rules to deduce conclusions Both work together to find proofs for complex geometry theorems.

English
8
43
258
54.5K
Matteo Palmonari retweetledi
TGDK
TGDK@tgdkjournal·
Issue 1 of Transactions on Graph Data & Knowledge (TGDK) published today: drops.dagstuhl.de/entities/issue…. TGDK is a new Diamond OA (no fees for authors/readers) journal about graphs for data & knowledge published by @dagstuhl, supported by SWSA. More about TGDK: dagstuhl.de/tgdk
TGDK tweet media
English
1
39
54
8.4K
Matteo Palmonari
Matteo Palmonari@MatteoPalmonari·
The position is about exploring novel AI models that keep humans in the loop to simplify data enrichment. Hot topics include techniques to improve LLM-based models with limited user feedback using active learning, transfer learning, in-prompt learning, and similar paradigms.
English
0
0
0
101
Matteo Palmonari
Matteo Palmonari@MatteoPalmonari·
@PMinervini 1) does the LLM learn inverse relations from data, across the multiple linguistic patterns in which two inverse relations appear? 2) is this knowledge used at inference time when asking questions flipping the relations?
English
0
0
1
34
Matteo Palmonari
Matteo Palmonari@MatteoPalmonari·
@PMinervini First, you conceptualize this in a better way than in the paper (I don't like the "A is B and B is A" formulation, but this may be due to my logic background). IMHO there are two issues at stake here:
English
1
0
1
43
Pasquale Minervini
Pasquale Minervini@PMinervini·
There is a cool new paper analysing the generalisation properties of LLMs (arxiv.org/abs/2308.03296) that claims they cannot capture higher-order patterns like “GW was the first US president” → “Who was the first US president? A: GW” -- I'm not sure that's the case tho 👇
Pasquale Minervini@PMinervini

@OwainEvans_UK Hi, I'm not sure that "they cannot" -- neural link predictors are essentially LMs over knowledge graph edges, and they can easily capture, e.g., "parentOf" is the inverse of "sonOf" (you can even extract/enforce them, as this ECML'17 paper of mine shows: aran.library.nuigalway.ie/handle/10379/7…)

English
3
5
32
16.1K
Debora Nozza
Debora Nozza@debora_nozza·
I'm thrilled and honored to have been awarded an #ERCStG for my project PERSONAE. I'll explore subjective #NLProc models from individuals' perspectives, with the goal of enhancing their efficiency and accessibility for all. Thanks to @ERC_Research for this opportunity!
Università Bocconi@Unibocconi

Great news as @ERC_Research assigns 7 Starting Grants to Bocconi young faculty plus 1 to an incoming researcher @debora_nozza @DanieleDurante2 @scottrw630 @NicolaLimodio @DeserrannoErika @michelacarlana #ERCStG #KnowledgeThatMatters knowledge.unibocconi.eu/notizia.php?id…

English
19
12
123
12.2K
Dr Owain Kenway
Dr Owain Kenway@owainkenway·
It's not an exhaustive test but I put the first few sentences of Silverman's book into ChatGPT and asked to complete the next paragraph and it invented something unrelated.
Dr Owain Kenway@owainkenway

E.g. "It must have been trained on my book because it can summarise my book" Or, it was trained on material that included articles that summarised your book. I'm also increasingly unconvinced that statistical data about word frequency is a derivative work and not just data.

English
3
0
3
499

Keşfet

@M2lSchool @unimib @FrancescoVisin @enrichmydata @FersiniE @ylecun @znz8 @FloydFabio