Pascal Malengrez (@PascalMlgz) header
Pascal Malengrez (@PascalMlgz) profile picture

Founder TAPE - GenAI Short form Video Messaging for business

Paris
Joined December 2012
Pascal Malengrez retweeted
The intelligence of humanity: $20/mo 1 Coffee: $6 We’ve got a lot of work to do!
98
65
9
1,420
445,865
Pascal Malengrez retweeted
AI won't grow your business. Experts using AI tools will.
6
6
34
4,052
Very good prompting vs fine tuning to use LLM on your dataset
There's a resurgence of interest in fine tuning LLMs I've yet to see a successful public use case where fine tuning > prompting. But here's where I see fine tuning *mattering*: First, fine tuning is for teaching an LLM specific tasks or behaviors Not teaching an LLM new knowledge. For new knowledge, use Retrieval (store your data in an outside database and strategically pull the right chunks in to give the LLM context to your question) But even in teaching LLMs specific tasks or behaviors - here's the catch... LLMs are remarkably good at picking up tasks and behaviors from just a good prompt THIS is what makes LLMs mind blowing after all So that begs the question. Where is fine tuning actually helpful? Some use cases I could see developing are teaching LLMs tasks that are exceptionally difficult to describe, or fit into ~10 examples you can add to a prompt. One way to think about this: if it would take someone a few weeks doing a task to 'master it' instead of being able to read training materials and get the picture... That *may* be a use case for fine tuning But proceed with caution To truly teach an LLM a new behavior or task, you'll need to treat this like a machine learning project, not just throwing examples in and getting magic in return (which it still blows my mind that ChatGPT does this so well for us). Things like: - Dataset design - Training and test data - Overfitting + more as the tooling around fine tuning gets more sophisticated The other obvious use case is cost. If you can get a super small language model to do a task instead of GPT-4, there's meaningful cost savings there. And if you're using a language model to do large scale tasks like triaging your customer support inbox, or analyzing public data for insights The costs can add up. But if you're wondering where the heck to invest in fine tuning... My answer at the moment for most businesses is still: Make sure you can't do it with prompts.
1
103
Pascal Malengrez retweeted
This is how many are feeling about A.I. these days.
16
11
3
87
12,317
True life advices for being good on video
I used to be shy & have no confidence. Now, I’ve built a YouTube channel with over 4,500,000 subscribers. Here are 5-hacks I use to be confident on camera:
38
Pascal Malengrez retweeted
seeing a lot of confusion about this, so for clarity: openai never trains on anything ever submitted to the api or uses that data to improve our models in any way.
493
438
195
5,396
1,364,390
WIP - new era for sure. And it is so exciting.
The future of running a business will be that you describe how you want it run, and it runs that way And that's crazy.
45
Your personal GenAI assistant on your phone featuring LLM, deep app links, chain of thoughts and full agency to get things done for you. Give it a try heydata.org/?twclid=23chj7g7…
43
Pascal Malengrez retweeted
The 10%: I think there are ocassions where arrogant/entitled founders can do well to *smash* incumbent markets to pieces, paving the way for innovation. There’s a few cases of that, where I’m not sure a well-balanced persona would have had the same impact. And yes, I’m older now, I try not to speak in absolutes, unless you’re my kid, then I’m always right.
1
3
127
Clear thoughts - all applied in what we’re building, not without couple of challenges
What I’ve learned, applied to AI market: DotCom veterans know that AI startups should explore new business models: synthetic data itself could be a product in its own right. Web2 veterans know that AI startups should deploy social features for user loyalty. Sharing Economy veterans know that AI startups need network effects to sustain or they will be left behind. Web3 veterans know that AI startups should not overly rely on other peoples platforms. I’ll speak on this at the next Llama Lounge event #AI
1
1
325
⁦@elonmusk⁩ this is a great opportunity for you to buy and put back in shape the Tesla of e-bikes. 🧵 theverge.com/2023/7/18/23798…
8
14
1
47
3,838
Please become an interested party and if you have not yet been approached the trustees, you can contact the trustees by sending an e-mail to [email protected]. The trustees will then inform you on the sales process.
1
1
5
397
Pascal Malengrez retweeted
A side hustle is a must. Replace your 9-5 by creating side hustles. I have prepared a ClickUp resource for side hustles. And for 24 hours, it's FREE 1. Follow me (so I can DM you) 2. RT this tweet 3. Reply "side"
889
636
9
742
161,839
Humor on AI at its best - let’s have some more of these!
You gotta hear this. I had the wildest dream last night about @framer. Something to do with them launching AI. framer.com/ai/ It was terrifying.
90
Likely future of B2B sales?
Ready for the day when your AI sales bot reaches out to my AI assistant bot, they're going to have a conversation and decide if your product is a good fit for my problems.
70
To every AI founder that is being told, you are a thin layer on top of commoditized models. ITA sold for $700M. Kayak sold for $2BN. Kayak was a thin layer on top of ITA. Just saying...
42
42
6
474
129,092