Cosimo Fedeli
@cosimofedeli.bsky.social
Tinkerer, data scientist, AI engineer, astrophysics PhD. I follow science, data, tech, business & finance. Views are my own.
https://www.linkedin.com/in/cosimo-fedeli
https://www.linkedin.com/in/cosimo-fedeli
Latest in a string of Financial Institutions, UBS appointed today a new Chief AI Officer. It was yesterday's news that Allianz UK also created a new Head of AI role.
As AI becomes more pervasive, firms are creating more structure around AI governance.
As AI becomes more pervasive, firms are creating more structure around AI governance.
UBS appoints chief AI officer
UBS is the latest banking giant to bring artificial intelligence into the C suite, tapping JP Morgan's Daniele Magazzeni as its first chief AI officer.
www.finextra.com
October 17, 2025 at 1:34 PM
Latest in a string of Financial Institutions, UBS appointed today a new Chief AI Officer. It was yesterday's news that Allianz UK also created a new Head of AI role.
As AI becomes more pervasive, firms are creating more structure around AI governance.
As AI becomes more pervasive, firms are creating more structure around AI governance.
It's interesting that amongst all the sound and fury about GenAI, sometimes some old-fashioned Machine Learning use-case appears in these deals between financial firms and tech companies.
Klarna and Google ink AI deal
Klarna has signed on to use Google Cloud's full AI stack, from infrastructure, to platform, to models.
www.finextra.com
October 13, 2025 at 9:45 AM
It's interesting that amongst all the sound and fury about GenAI, sometimes some old-fashioned Machine Learning use-case appears in these deals between financial firms and tech companies.
A nice read, with some definitely relatable points, particularly data fragmentation and infrastructural challenges.
Model management, integration, and explainability remain key constraints.
Model management, integration, and explainability remain key constraints.
Why Financial Services Are Betting Big on AI-Powered Analytics (And What Most Get Wrong): By Sergiy Fitsak
Walk into any bank's strategy meeting today and you'll hear the same refrain: "We need AI in
www.finextra.com
October 6, 2025 at 7:09 AM
A nice read, with some definitely relatable points, particularly data fragmentation and infrastructural challenges.
Model management, integration, and explainability remain key constraints.
Model management, integration, and explainability remain key constraints.
This was only a matter of time.
I have little doubt GenAI will be transformative for some tasks and applications, but the work needs to be put in to craft a solution that integrates with, and improves upon, existing business processes.
I have little doubt GenAI will be transformative for some tasks and applications, but the work needs to be put in to craft a solution that integrates with, and improves upon, existing business processes.
Sure, AI may be an environmental disaster and produce mid outcomes in cases where it doesn't straight up lie, but at least it's extremely unprofitable.
As Klarna flips from AI-first to hiring people again, a new landmark survey reveals most AI projects fail to deliver
Just 1 in 4 AI investments bring in the ROI they promise—but CEOs just can’t resist the technology.
fortune.com
May 13, 2025 at 9:27 PM
This was only a matter of time.
I have little doubt GenAI will be transformative for some tasks and applications, but the work needs to be put in to craft a solution that integrates with, and improves upon, existing business processes.
I have little doubt GenAI will be transformative for some tasks and applications, but the work needs to be put in to craft a solution that integrates with, and improves upon, existing business processes.
A nice summary of Oracle's annual event for industry analysts, where LLMs and Agentic AI were discussed. I'm glad to be part of the team that brings these transformative technologies to the Financial Services space.
Innovation in Risk Management and Compliance
What does it look like when a leading provider of Risk software pivots to innovation? I got a good answer to this question when I attended Oracle’s annual event for industry analysts held at the Oracl...
www.celent.com
May 13, 2025 at 3:27 PM
A nice summary of Oracle's annual event for industry analysts, where LLMs and Agentic AI were discussed. I'm glad to be part of the team that brings these transformative technologies to the Financial Services space.
🧪 One of the most profound concepts in physics is that of least action, whereby objects follow trajectories that minimize specific functionals. One manifestation of this is Fermat’s principle, which dictates that light rays follow the path between two points that minimizes light’s travel time
April 2, 2025 at 4:10 PM
🧪 One of the most profound concepts in physics is that of least action, whereby objects follow trajectories that minimize specific functionals. One manifestation of this is Fermat’s principle, which dictates that light rays follow the path between two points that minimizes light’s travel time
I’ve been reading this interesting paper, where the authors propose a generalization of the Tree of Thought prompting framework to improve the reasoning capabilities of AI Agents
April 1, 2025 at 3:36 PM
I’ve been reading this interesting paper, where the authors propose a generalization of the Tree of Thought prompting framework to improve the reasoning capabilities of AI Agents
Last week I had a chat with students from Milan's Polytechnic school about career, AI, and Data Science.
It's always interesting to speak with people who are one full generation younger than yourself. You learn from them as much as they learn from you.
It's always interesting to speak with people who are one full generation younger than yourself. You learn from them as much as they learn from you.
March 31, 2025 at 10:19 AM
Last week I had a chat with students from Milan's Polytechnic school about career, AI, and Data Science.
It's always interesting to speak with people who are one full generation younger than yourself. You learn from them as much as they learn from you.
It's always interesting to speak with people who are one full generation younger than yourself. You learn from them as much as they learn from you.
On some personal news, I'm happy to share that the Cloud service myself and a great team at Oracle have been working on for the past year has been released.
We're leveraging Generative AI Agents to automate Financial Crime investigations and analysis workflows.
Much still to be done!
We're leveraging Generative AI Agents to automate Financial Crime investigations and analysis workflows.
Much still to be done!
Oracle Brings AI Agents to the Fight Against Financial Crime
/PRNewswire/ -- Oracle Financial Services is supercharging its Investigation Hub Cloud Service with the addition of a broad class of AI agents and agentic...
www.prnewswire.com
March 13, 2025 at 4:51 PM
On some personal news, I'm happy to share that the Cloud service myself and a great team at Oracle have been working on for the past year has been released.
We're leveraging Generative AI Agents to automate Financial Crime investigations and analysis workflows.
Much still to be done!
We're leveraging Generative AI Agents to automate Financial Crime investigations and analysis workflows.
Much still to be done!
What struck me most when I was first reading about trade-based money laundering was that if you're clever (and dishonest) enough with your invoicing, you can move money around without actually moving money.
www.finextra.com/the-long-rea...
www.finextra.com/the-long-rea...
March 5, 2025 at 5:09 PM
What struck me most when I was first reading about trade-based money laundering was that if you're clever (and dishonest) enough with your invoicing, you can move money around without actually moving money.
www.finextra.com/the-long-rea...
www.finextra.com/the-long-rea...
Reposted by Cosimo Fedeli
This looks cool; the TreeSatAI-Time-Series enhances tree species classification with a full year of Sentinel-1 & -2 data! Perfect for AI-driven forest monitoring & biodiversity research. #RemoteSensing #TreeMapping #MachineLearning - huggingface.co/datasets/IGN...
February 28, 2025 at 3:38 PM
This looks cool; the TreeSatAI-Time-Series enhances tree species classification with a full year of Sentinel-1 & -2 data! Perfect for AI-driven forest monitoring & biodiversity research. #RemoteSensing #TreeMapping #MachineLearning - huggingface.co/datasets/IGN...
Working in AI and Financial Crime & Compliance myself, this resonated quite a bit. We're still very much learning what LLM Agents can actually do, so it's not surprising that there are diverging opinions on a matter as delicate as Regulatory Compliance.
Compliance professionals split over value of AI
Growing levels of financial crime and regulatory scrutiny are forcing organisations to step up spending on compliance, but they are less sold on AI as the standalone solution, according to a global su...
www.finextra.com
February 24, 2025 at 1:50 PM
Working in AI and Financial Crime & Compliance myself, this resonated quite a bit. We're still very much learning what LLM Agents can actually do, so it's not surprising that there are diverging opinions on a matter as delicate as Regulatory Compliance.
Reposted by Cosimo Fedeli
"Mathematics is the queen of the sciences and number theory is the queen of mathematics. She often condescends to render service to astronomy and other natural sciences, but in all relations, she is entitled to first rank."
— Carl F. Gauss, who passed away on 23 Feb. 1855
🔭 🧪 ⚛️ #mathematics 1/2
— Carl F. Gauss, who passed away on 23 Feb. 1855
🔭 🧪 ⚛️ #mathematics 1/2
February 23, 2025 at 8:24 PM
"Mathematics is the queen of the sciences and number theory is the queen of mathematics. She often condescends to render service to astronomy and other natural sciences, but in all relations, she is entitled to first rank."
— Carl F. Gauss, who passed away on 23 Feb. 1855
🔭 🧪 ⚛️ #mathematics 1/2
— Carl F. Gauss, who passed away on 23 Feb. 1855
🔭 🧪 ⚛️ #mathematics 1/2
Reposted by Cosimo Fedeli
I've been thinking a lot about "slow science" recently, about pushing against the breakneck pace (especially in data science/ML). Feel like I needed to read this.
#academicsky 🧪 ⚗️ #compchem
#academicsky 🧪 ⚗️ #compchem
Michael Crichton (Jurassic Park) has some thoughts that feel very relevant today:
February 21, 2025 at 7:52 PM
I've been thinking a lot about "slow science" recently, about pushing against the breakneck pace (especially in data science/ML). Feel like I needed to read this.
#academicsky 🧪 ⚗️ #compchem
#academicsky 🧪 ⚗️ #compchem
I still don't get why people are so hung up with the fact that LLMs are bad at math. LLMs are language models. They're not supposed to know how to make multiplications, if not to the extent that multiplications can be found in the training corpus.
you fucked up a perfectly good computer is what you did. look at it. it's got innumeracy
February 13, 2025 at 10:29 PM
I still don't get why people are so hung up with the fact that LLMs are bad at math. LLMs are language models. They're not supposed to know how to make multiplications, if not to the extent that multiplications can be found in the training corpus.
I couldn't agree more with this. Reading has immeasurable benefits for both the individual and the community.
I get that a lot of people don’t like to read, which encourages reliance on ChatGPT and TikTok - but “aversion to reading” has been an issue for basically forever, and I believe that instead of just giving up, we should still be strongly encouraging people to read, because it’s good for society
February 13, 2025 at 10:17 PM
I couldn't agree more with this. Reading has immeasurable benefits for both the individual and the community.
There was a lot of discussion last week about the so-called “unlimited context length” for some of the LLMs that have been released by DeepSeek. The concept is explained quite nicely in this paper, that I summarize below. I’m hoping this helps make some clarity. 🧵
Long Context Compression with Activation Beacon
Long context compression is a critical research problem due to its significance in reducing the high computational and memory costs associated with LLMs. In this paper, we propose Activation Beacon, a...
arxiv.org
February 9, 2025 at 9:33 AM
There was a lot of discussion last week about the so-called “unlimited context length” for some of the LLMs that have been released by DeepSeek. The concept is explained quite nicely in this paper, that I summarize below. I’m hoping this helps make some clarity. 🧵
The first experimental evidence of relativistic time dilation I came into contact with, so many years ago.
Muons
xkcd.com
February 8, 2025 at 8:33 PM
The first experimental evidence of relativistic time dilation I came into contact with, so many years ago.
A recent publication shows that small (but numerous) methane emitters across oil and gas operations are responsible for the majority of global emissions.
Study shows small sources account for large share of oil & gas methane emissions, underscores importance of key federal initiatives
For a dozen years, study after scientific study has documented the vast amount of methane released to the atmosphere by oil and gas operations worldwide. Faced with mounting data, both producers and r...
energycentral.com
February 6, 2025 at 10:25 PM
A recent publication shows that small (but numerous) methane emitters across oil and gas operations are responsible for the majority of global emissions.
I’ve been reading this interesting paper, where the authors propose a generalization of the Tree of Thought prompting approach to improve the reasoning capabilities of AI Agents.
Graph of Thoughts: Solving Elaborate Problems with Large Language Models
We introduce Graph of Thoughts (GoT): a framework that advances prompting capabilities in large language models (LLMs) beyond those offered by paradigms such as Chain-of-Thought or Tree of Thoughts (T...
arxiv.org
February 3, 2025 at 5:47 PM
I’ve been reading this interesting paper, where the authors propose a generalization of the Tree of Thought prompting approach to improve the reasoning capabilities of AI Agents.
Hence, Karl Popper's principle of falsifiability.
"That is how we do science: We find something, then we challenge every aspect of it to ensure we’re not seeing something we want to see rather than what is actually there."
Inside the hunt for habitable planets: 🧪
Inside the hunt for habitable planets: 🧪
Discovering the First Other Earths
Inside the hunt for habitable planets.
nautil.us
February 2, 2025 at 9:09 PM
Hence, Karl Popper's principle of falsifiability.
It's refreshing to occasionally read about startups that didn't make it, rather than the successful ones. There are lessons to be learned from failure, maybe more so than from success.
Fintech startup Cushion shuts down after 8 years and over $20 million in funding
Fintech startup Cushion shuts down after 8 years and over $20 million in funding
Cushion, a fintech startup that described itself as the “Plaid for buy now, pay later (BNPL),” has shut down. On Thursday, founder and CEO Paul Kesserwani posted on LinkedIn about the decision to wind down the company at the end of 2024. In the post,…
tcrn.ch
January 30, 2025 at 10:15 PM
It's refreshing to occasionally read about startups that didn't make it, rather than the successful ones. There are lessons to be learned from failure, maybe more so than from success.
DeepSeek really played a practical joke on AI companies. Curious to see how this will evolve.
LIVE: Tech stocks sell off as DeepSeek fears hit Wall Street.
What questions do you have? Join our Live Q&A
What questions do you have? Join our Live Q&A
Live Q&A: Tech Stocks Tumble as DeepSeek Fears Hit Wall Street
Bloomberg journalists answer your questions about impact of DeepSeek on US tech stocks
www.bloomberg.com
January 27, 2025 at 6:47 PM
DeepSeek really played a practical joke on AI companies. Curious to see how this will evolve.
Interesting learnings from HSBC's decision to close its international payment app after less than one year from launch.
What resonated most: "Innovation for innovation’s sake doesn’t work: if you’re not solving real customer pain points with innovative solutions, your product is destined to fail."
What resonated most: "Innovation for innovation’s sake doesn’t work: if you’re not solving real customer pain points with innovative solutions, your product is destined to fail."
HSBC’s Zing Shutdown: What Traditional Banks Can Learn from FinTech Disruption: By Ritesh Jain
The closure of Zing, HSBC's international payments app launched just last year, is both disappointin...
www.finextra.com
January 27, 2025 at 4:34 PM
Interesting learnings from HSBC's decision to close its international payment app after less than one year from launch.
What resonated most: "Innovation for innovation’s sake doesn’t work: if you’re not solving real customer pain points with innovative solutions, your product is destined to fail."
What resonated most: "Innovation for innovation’s sake doesn’t work: if you’re not solving real customer pain points with innovative solutions, your product is destined to fail."
Something that will need to eventually be established as our understanding of GenAI improves is whether foundational LLMs have inference capabilities.
IDEA: Enhancing the Rule Learning Ability of Large Language Model Agent through Induction, Deduction, and Abduction
While large language models (LLMs) have been thoroughly evaluated for deductive and inductive reasoning, their proficiency in holistic rule learning in interactive environments remains less explored. ...
arxiv.org
January 27, 2025 at 11:04 AM
Something that will need to eventually be established as our understanding of GenAI improves is whether foundational LLMs have inference capabilities.