High Signal Podcasts Evidence ledger
Method
Browse
How the evidence works

Public evidence record

Alessio Fanelli

Partner · Decibel

Claims
224
Episodes
65
Shows
1
Named items
3

Books, apps, and tools

The evidenced stack.

Browse the grouped index →

tool / uses

Zo

“I think there’s been one thing, I use another thing called zo, which is kinda like a cloud computer plus agent.”

Latent Space · 17 Mar 2026

Evidence receipt · Source ↗

tool / likes

VCR

“First of all, you have very good support for mocking in unit tests, which is something that a lot of other frameworks don't do. So, you know, my favorite Ruby library is VCR because it just, you know, it just lets me store the HTTP requests and replay them.”

Latent Space · 6 Feb 2025

Evidence receipt · Source ↗

other / likes

Simulative AI

“Yeah, I'm a big fan of Simulative AI. We had a summer of Simulative AI. Another term we're trying to coin.”

Latent Space · 27 Sept 2024

Evidence receipt · Source ↗

Claim ledger

What Alessio said.

43 transcript-backed records

01 / evaluation

I was gonna say, to me, driving feels like a great next token prediction thing because you’re kinda like on a path and like, it doesn’t really matter what you’ve done before.

“I was gonna say, to me, driving feels like a great next token prediction thing because you’re kinda like on a path and like, it doesn’t really matter what you’ve done before.”
Speaker
Alessio Fanelli
Publisher
Latent Space

02 / evaluation

The problem is that you need twice the amount of ram, twice the amount of, you know, it’s like, it’s kind of taxing on the machine.

“The problem is that you need twice the amount of ram, twice the amount of, you know, it’s like, it’s kind of taxing on the machine.”
Speaker
Alessio Fanelli
Publisher
Latent Space

06 / evaluation

Yeah. And the big thing was like the mixed trial price fights, you know, and I think now it's almost like there's nowhere to go, like, you know, Gemini Flash is like basically giving it away for free.

“Yeah. And the big thing was like the mixed trial price fights, you know, and I think now it's almost like there's nowhere to go, like, you know, Gemini Flash is like basically giving it away for free.”
Speaker
Alessio Fanelli
Publisher
Latent Space

07 / evaluation

Um, and I think today the, the problem is that, Yeah, the agents are, that most people are building are good at following instruction, but are not as good as like extracting them from you.

“Um, and I think today the, the problem is that, Yeah, the agents are, that most people are building are good at following instruction, but are not as good as like extracting them from you.”
Speaker
Alessio Fanelli
Publisher
Latent Space

08 / evaluation

I think to me that the most interesting is like rest and GraphQL is almost more interesting in the world of agents because agents could come up with so many different things to query versus like before I always thought GraphQL was kind of like not really necessary because like, you know what you need, just build the rest end point for it.

“I think to me that the most interesting is like rest and GraphQL is almost more interesting in the world of agents because agents could come up with so many different things to query versus like before I always thought GraphQL was kind of like not really necessary because like, you know what you need, just build the rest end point for it.”
Speaker
Alessio Fanelli
Publisher
Latent Space

12 / evaluation

And eventually, you know, I started doing venture six, five years ago. And I think just like so many people in Europe reach out and ask, hey, can you like talk to our team and they just cannot comprehend like the risk appetite that people have here.

“And eventually, you know, I started doing venture six, five years ago. And I think just like so many people in Europe reach out and ask, hey, can you like talk to our team and they just cannot comprehend like the risk appetite that people have here.”
Speaker
Alessio Fanelli
Publisher
Latent Space

13 / evaluation

When I tried to set up Slack, it was like, hey, give me access to all channels and everything, which for the average person probably makes sense because you don't want to re-prompt them every time you add new channels.

“When I tried to set up Slack, it was like, hey, give me access to all channels and everything, which for the average person probably makes sense because you don't want to re-prompt them every time you add new channels.”
Speaker
Alessio Fanelli
Publisher
Latent Space

14 / evaluation

I've written a post called Maximum Enterprise Utilization, kind of like you have MFU for GPUs, but it's basically like so many people are focused on, oh, it's going to like displace jobs and whatnot. But I'm like, there's so much work that people don't do because they don't have the people.

“I've written a post called Maximum Enterprise Utilization, kind of like you have MFU for GPUs, but it's basically like so many people are focused on, oh, it's going to like displace jobs and whatnot. But I'm like, there's so much work that people don't do because they don't have the people.”
Speaker
Alessio Fanelli
Publisher
Latent Space

18 / evaluation

It's not, it's not what gives us our edge, but it certainly means that then we don't have to build it and maintain it afterwards. So, it's a really good first step, I think, in, like, the overall maturity of the fine tuning product and API in terms of where they're going to see those early products.

“It's not, it's not what gives us our edge, but it certainly means that then we don't have to build it and maintain it afterwards. So, it's a really good first step, I think, in, like, the overall maturity of the fine tuning product and API in terms of where they're going to see those early products.”
Speaker
Alessio Fanelli
Publisher
Latent Space

20 / evaluation

I think there was obviously the Kepler, and then there was Chinchilla, and then people kind of got the Llama scaling law, like the 100 to 200x parameter to token ratio.

“I think there was obviously the Kepler, and then there was Chinchilla, and then people kind of got the Llama scaling law, like the 100 to 200x parameter to token ratio.”
Speaker
Alessio Fanelli
Publisher
Latent Space

21 / evaluation

Are people just finding out recently about these problems because now the scores are getting so high that you're actually inspecting the benchmarks and maybe in the past you were scoring so badly that maybe you weren't as worried about the overall quality?

“Are people just finding out recently about these problems because now the scores are getting so high that you're actually inspecting the benchmarks and maybe in the past you were scoring so badly that maybe you weren't as worried about the overall quality?”
Speaker
Alessio Fanelli
Publisher
Latent Space

23 / evaluation

For example, with one system, there are like 780 endpoints. And if you're actually trying to do vector similarity, it's not that good because the people that wrote the specs didn't have in mind making them like semantically apart.

“For example, with one system, there are like 780 endpoints. And if you're actually trying to do vector similarity, it's not that good because the people that wrote the specs didn't have in mind making them like semantically apart.”
Speaker
Alessio Fanelli
Publisher
Latent Space

24 / evaluation

People were like, okay, this is better than this on this benchmark, blah, blah, blah, because maybe they did not have a lot of use cases that they did frequently.

“People were like, okay, this is better than this on this benchmark, blah, blah, blah, because maybe they did not have a lot of use cases that they did frequently.”
Speaker
Alessio Fanelli
Publisher
Latent Space

25 / evaluation

Cloud is better. It's very good, you know, it's much better, it seems to me, it's much better than GPT 4 at doing writing that is more, you know, I don't know, it just got good vibes, you know, like the GPT 4 text, you can tell it's like GPT 4, you know, it's like, it always uses certain types of words and phrases and, you know, maybe it's just me because I've now done it for, you know, So, I've read like 75, 80 generations of these things next to each other.

“Cloud is better. It's very good, you know, it's much better, it seems to me, it's much better than GPT 4 at doing writing that is more, you know, I don't know, it just got good vibes, you know, like the GPT 4 text, you can tell it's like GPT 4, you know, it's like, it always uses certain types of words and phrases and, you know, maybe it's just me because I've now done it for, you know, So, I've read like 75, 80 generations of these things next to each other.”
Speaker
Alessio Fanelli
Publisher
Latent Space

26 / evaluation

I think before we wrap, you have written a blog post that can show about good hearts law impact in ML, which is, you know, when you measure something, then the thing that you measure is not a good metric anymore because people optimize for it.

“I think before we wrap, you have written a blog post that can show about good hearts law impact in ML, which is, you know, when you measure something, then the thing that you measure is not a good metric anymore because people optimize for it.”
Speaker
Alessio Fanelli
Publisher
Latent Space

27 / evaluation

I think if anything, RAG's complexity goes up and up the more you use it, you know, because you have more data sources, more things you want to put in there.

“I think if anything, RAG's complexity goes up and up the more you use it, you know, because you have more data sources, more things you want to put in there.”
Speaker
Alessio Fanelli
Publisher
Latent Space

28 / evaluation

I think in Europe, I walked through a lot of the posters and whatnot, there seems to be mode collapse in a way in the research, a lot of people working on the same things.

“I think in Europe, I walked through a lot of the posters and whatnot, there seems to be mode collapse in a way in the research, a lot of people working on the same things.”
Speaker
Alessio Fanelli
Publisher
Latent Space

30 / evaluation

I know we kind of binned the lightning round in the last few episodes, but I think for you two, one of the questions we used to ask is like, what's the most interesting unsolved question in AI?

“I know we kind of binned the lightning round in the last few episodes, but I think for you two, one of the questions we used to ask is like, what's the most interesting unsolved question in AI?”
Speaker
Alessio Fanelli
Publisher
Latent Space

33 / evaluation

I think, to me, the biggest takeaway was like and I was talking with Mike Conover, another friend of the podcast, about this is they're kind of staying in the single threaded, like, synchronous use cases lane, you know?

“I think, to me, the biggest takeaway was like and I was talking with Mike Conover, another friend of the podcast, about this is they're kind of staying in the single threaded, like, synchronous use cases lane, you know?”
Speaker
Alessio Fanelli
Publisher
Latent Space

34 / evaluation

So I think the easiest thing for people to grasp so far has been Mojo, which is a superset of Python. And I think everybody talks about that because it's easier to grasp, but Modular's goal is to build a unified AI engine.

“So I think the easiest thing for people to grasp so far has been Mojo, which is a superset of Python. And I think everybody talks about that because it's easier to grasp, but Modular's goal is to build a unified AI engine.”
Speaker
Alessio Fanelli
Publisher
Latent Space

35 / evaluation

I think in one of your previous podcasts, you mentioned leaving people behind, you know, that are like not experts in certain things and they can't contribute.

“I think in one of your previous podcasts, you mentioned leaving people behind, you know, that are like not experts in certain things and they can't contribute.”
Speaker
Alessio Fanelli
Publisher
Latent Space

37 / evaluation

If you have a eight bits model quantized down, you need one byte per parameter. So for example, in an H100, which is 80 gigabyte of memory, you could fit a 70 billion parameters in eight, you cannot fit a FP32 because you will need like 280 gigabytes of memory.

“If you have a eight bits model quantized down, you need one byte per parameter. So for example, in an H100, which is 80 gigabyte of memory, you could fit a 70 billion parameters in eight, you cannot fit a FP32 because you will need like 280 gigabytes of memory.”
Speaker
Alessio Fanelli
Publisher
Latent Space

38 / evaluation

I think I've talked about this on the podcast, but this idea of like just-in-time UIs, you know, like each type of user wants to interact in a different way.

“I think I've talked about this on the podcast, but this idea of like just-in-time UIs, you know, like each type of user wants to interact in a different way.”
Speaker
Alessio Fanelli
Publisher
Latent Space

39 / evaluation

I think I talked about it on the podcast before, but like the switch from syntax to like semantics, like developers used to be focused on the syntax and not the meaning of what they're writing.

“I think I talked about it on the podcast before, but like the switch from syntax to like semantics, like developers used to be focused on the syntax and not the meaning of what they're writing.”
Speaker
Alessio Fanelli
Publisher
Latent Space

40 / evaluation

I'm building an agent internally for us. And Guardrails are obviously very exciting because once you set the initial prompt, like the model creates its own prompts.

“I'm building an agent internally for us. And Guardrails are obviously very exciting because once you set the initial prompt, like the model creates its own prompts.”
Speaker
Alessio Fanelli
Publisher
Latent Space

41 / evaluation

Yeah. This is super interesting because right now a lot of products are kind of the same because all I do is they call it the model and some are prompted a little differently, but you can only guess so much delta between them in the future.

“Yeah. This is super interesting because right now a lot of products are kind of the same because all I do is they call it the model and some are prompted a little differently, but you can only guess so much delta between them in the future.”
Speaker
Alessio Fanelli
Publisher
Latent Space

43 / evaluation

The question is, if you wanna compete against these companies, maybe the model is not what you're gonna do it with because the open source kind of commoditizes it.

“The question is, if you wanna compete against these companies, maybe the model is not what you're gonna do it with because the open source kind of commoditizes it.”
Speaker
Alessio Fanelli
Publisher
Latent Space
Search evidence