Mathieu Fortier

Director, ML/#nlproc @Coveo | NLP Science & Engineering | R&D Leadership | Views expressed here are my own

[François Chollet @fchollet]

This is what happens when you try to use a LLM as a store of information.

[Isabel Angelo @IsabelNAngelo]

Unfortunately a simple google search would tell us that JWST actually did not "take the very first picture of a planet outside of our own solar system" and this is literally in the ad for Bard so I wouldn't trust it yet

[Mathieu Fortier @mathieufortier]

@IsabelNAngelo “Whoopsies”

[Mathieu Fortier @mathieufortier]

@FlaxSearch @jobergum @HanchungLee We feel the same pain I see! But let’s not be pessimistic here: LLMs can bring very tangible value in that equation. It’s a harder problem, sure, but that’s what gets me fired up. I’ll keep trying :)

[Mathieu Fortier @mathieufortier]

@FlaxSearch @jobergum Exactly - That’s what a lot of outsiders don’t understand about enterprise search and knowledge management: Security and confidentiality are paramount.

[Mathieu Fortier @mathieufortier]

@jobergum Sexy!

[Mathieu Fortier @mathieufortier]

@lateinteraction 🔥

[Mathieu Fortier @mathieufortier]

Maybe "Confidence is all you need"?

From https://t.co/W5RccFu1uB (great read!)

#llm #nlproc https://t.co/TUHVVpzFIQ

[Mathieu Fortier @mathieufortier]

It's good to embrace advances and be excited about it, but it's probably best not to let the robots roam free too much - They still require their human masters (for now). https://t.co/iKIjikr3TN

[Mathieu Fortier @mathieufortier]

@jobergum It is part of the vision we are building towards at @coveo. Having years of experience with enterprise search is key. For example those who think “all you need to disrupt is a fancy LLM” are in for a rude awakening…

[Mathieu Fortier @mathieufortier]

@jobergum Plus we all know encoding at the document level is all you need to produce great dense retrieval results… /s

[Mathieu Fortier @mathieufortier]

@jobergum Funny, that’s a discussion I’ve also had (more than once) in recent years. On a more philosophical note, I’d say we could all use a bit less speed in our lives. And in the pace of new AI research. Would be nice to feel all caught up once in a while…

[Brendan O'Donoghue @bodonoghue85]

If we view LLMs in an RL way, then outputs are just rollouts from the policy. The 'prompt' is just the initial state, it encodes the starting condition of the agent and, implicitly, a goal. Prompt engineering == finding a good initial state for the agent to achieve your goal.

[Mathieu Fortier @mathieufortier]

Check out my latest article: Humble Thoughts on ChatGPT https://t.co/VnQ4zCeLtP via @LinkedIn

[Mathieu Fortier @mathieufortier]

What I think is the next frontier from there? Access to nonparametric memory (DBs, KGs… The whole Internet?) to bridge the gap between static (e.g.: laws of Physics) and dynamic (e.g.: who’s president of X?) World Knowledge.

[Mathieu Fortier @mathieufortier]

I find the mountain of “X is dead” posts about ChatGPT pretty amusing. Sure, it’ll be great to regurgitate an essay on past History. Now ask it to link those ideas and concepts to very recent and ongoing world events.

P.S.: Still bullish we’ll solve for that eventually.

[Mathieu Fortier @mathieufortier]

@jobergum Agreed, promptagator is really 🤯 stuff! To play devil’s advocate a bit, I’m always quite cautious when considering IRL / Enterprise use of generative augmentation approaches due to introduction of unforeseen and undesirable biases. It got my attention though…

[EigenGender 🔸 ✈️ sf Aug 29- Sep 12 @EigenGender]

Universities are really not prepared for the flood of LLM-based honor code violations that are coming, and I'm really not sure what it's possible to do about it. I'd estimate ~5% of university students will submit an assignment largely written by a LLM by end of 2023.

[Mathieu Fortier @mathieufortier]

@shreyas @coveo is. Fun place to work at, many exciting challenges, great culture. https://t.co/ShmyRtMkSa