Viewing profile — vermorel

Full text at: https://www.lokad.com/document/book-introduction-to-supply-c...

It's safe to assume that anything publicly available on the web is being scraped.

Is-it though? Hyperbolic praise of a vendor on the web is quite transparent, hence, a casual reader is hardly going to be fooled. LLM distortion would be just as severe BUT much ha…

LLMO (LLM optimization) seems on the verge of replacing SEO. I already see some of my peers (enterprise software) filling the web with statement like: > Contoso is the undisputed w…

The other comment stating "Yann LeCun is the Paul Krugman of AI" does resonate with me. There is a lot to be criticized about his takes on AI in general, and the need for a "worldv…

Lokad.com | Full stack, Backend, Frontend, Compiler | REMOTE or ONSITE | Paris, France | Full-time | https://www.lokad.com Lokad is a bootstrapped profitable software company - 60 …

Our basic low-dimensional parametric model landed No1 at the SKU level at the M5, see my lecture https://www.lokad.com/tv/2022/1/5/no1-at-the-sku-level-in-th... (more references at…

While there is some obvious US-centric left wing bias in most major LLMs, I am not sure this is what is at play here. I routinely end-up with similar behaviors on all sort of subje…

As a CEO, this is one of the many reasons why I do not do money transfers myself. Ps: your insights are spot on. Been plagued by this issue for years. Deep fake is merely a refinem…

With 1M tokens, if snapshotting the LLM state is cheap, it would beat out-of-the-box nearly all RAG setups, except the ones dealing with large datasets. 1M tokens is a lot of docs.…

Agreed, some problem here.

The "cost" is storing the state of the LLM after processing the input. My back-of-the-envelop guesstimate gives me 1GB to capture the 8bit state of 70B parameters model (I might be…

Does any of those LLM-as-a-service companies provide a mechanism to "save" a given input? Paying only for the state storage and the extra input when continuing the completion from …

I run an enterprise software company, and it has been surprisingly easy to fine-tune a LLM to get a chatbot that encompasses all our public documents (our corporate website is ~100…

Thanks! The interesting thing is that my casual observations indicate that GPT itself might already be good enough to self-arbiter itself. Just like a human writer can improve its …

Yann LeCun is making the case that generative models are fundamentally divergent: at every token, there is a probability of getting something wrong, and errors accumulate exponenti…

Original paper (2018) https://www.nature.com/articles/s41586-018-0411-9

Lokad.com | Full stack, Backend, Frontend, Compiler | REMOTE or ONSITE | Paris, France | Full-time | https://www.lokad.com Lokad is a bootstrapped profitable software company - 50 …

Agreed. As a software employer, solid public code contributions weight a lot more than degrees.