Prediction Not checkable as stated
Pesenti: AI matching human intelligence is at least 20-30 years away
“And I don't believe anybody has a good view as to when will match human intelligence, but it's not going to happen in the next decade, not going to happen in the next two or three decades. It's going to take much longer.”
Prediction Not checkable as stated
Pesenti in 2020: AI will not become accessible to non-programmers anytime soon
“I don't buy that AI will be accessible to people who don't have a programming background anytime soon, right? So, but it will be part of, you know, machine learning will be part of the tool set of Any programmers, but it would still be, you know, for kind of m…”
Prediction Not checkable as stated
Pesenti: Deepfake detection will never be a solved problem
“It's never going to be a soft problem, right? So it's always going to be an arm race. I'm confident that we can actually identify part of it. I think we can act on it. We can label it in the case where it matters to users, but I'm also confident that it's goin…”
Prediction Not checkable as stated
Pesenti: AI will not enter another winter anytime soon
“In my opinion, you know, we're not going to go to any AI winter anytime soon.”
Opinion
Pesenti in 2020: Large neural chatbots are not ready for consumer products
“It's not ready for prime consumption, let's be clear. You know, like these bots have lots of interesting side effects. You know, you cannot really control what they're gonna say. But they are quite entertaining, quite engaging, and a lot more human-like than y…”
Prediction Held up
Pesenti in 2020: 10x annual AI compute scaling is economically unsustainable
“So we're still pushing hard, but it's not going to be 10 X per year. It's just not sustainable.”
Assertion Supported
Pesenti in 2020: Multi-million-dollar AI training runs are unsustainable for Facebook
“Where it becomes millions is when you do training runs. So some of the training runs in the most advanced system that it comes from our company or other companies out there are starting to be extremely expensive. Yeah. Like you can look at one run in the scale…”
Insight
Pesenti: General intelligence does not exist, as human intelligence is domain-specific
“Human intelligence is no, it's not general. It's actually very, very specific to our world. It's very biased. It's really customized to survival human in, on our planet. So the concept of general intelligence, I mean, yes, humans tend to have a more general in…”
Prediction Not checkable as stated
Pesenti: Self-supervised learning will supersede other AI techniques across fields
“Self-supervised learning is really showing promises everywhere, you know, so it's obviously that's the technique behind large language models, but our, you know, my team came up with new papers around vision. So we believe actually that these techniques will s…”
Assertion Not checkable as stated
Two AI-generated ALS compounds outperformed standard-of-care treatments in lab validation
“We actually put the five hypothesis in the lab, and at this point, it's in vitro and in vivo validation, and, you know, two of them didn't work well one of them work as well as this kind of standard of care at the moment, and two of them work actually much bet…”
Assertion Contradicted
Pesenti: Facebook employs 30,000 content moderators
“So we have actually 30,000 moderators that try to really understand if the content that's published on a platform satisfy our policy, but 30,000, given the number, the amount of content, and you're talking about billions of pieces of content every day is not e…”
Assertion Partly supported
Pesenti in 2020: 9B parameter Blenderbot is the largest chat model
“It's the largest chat model out there. It's nine billion parameters.”
Disclosure
Pesenti: Inference remains Facebook's largest machine learning cost
“Actually, To be clear, the most costly thing we do in ML is still inference cost, because when you put a piece of content within Facebook, it's running hundreds of different ML based algorithm, and they all run on machine parallel, and it's using a huge number…”
Assertion Supported
Pesenti: Facebook AI transformer models matched or beat Mathematica at math
“My team also applied transformer models to mathematics, you know, doing, like, things like partial derivation equation or integration and showing that, hey, it could work as well better than Mathematica or some, you know, hundred page long algorithm.”
Disclosure
Pesenti: Facebook is going end-to-end all-in on PyTorch
“We're definitely going all in as PyTorch, you know, end to end. So I think initially when we launched the Onyx strategy, it was more like a multi-framework world. And we had actually two framework internally between PyTorch and Cafe Two, but we're still suppor…”
Assertion Not checkable as stated
The overall cost of scientific discovery may be increasing exponentially
“Because the number of researchers is almost increasing exponentially, but our growth rate actually is kind of almost decreasing, it may be true not just for drugs, but, you know, overall for humanity that the cost of discovery is increasing exponentially.”
Disclosure
BenevolentAI has initiated a Phase II clinical trial
“We actually started just now. A clinical trial phase two, so we're going all the way from, ah, the AI to the actual clinical trials.”
Opinion
Chemists consider molecules from published AI generative chemistry models impractical
“Now there are lots of great papers that have been written around using machine learning and neural network around generating new chemistry, but if you really look at them, if you actually get chemists to look at what's generated, they will say, ah, that's a jo…”
Insight
Large pharma relies on acquisitions rather than internal drug research
“Large pharma is not really the research engine anymore, and that a lot of the large pharma actually will acquire more than develop themselves.”
Insight
Large pharma incentives encourage researchers to push failing programs to completion
“If you're in a large pharma, and you're just in one area, your goal is to push your programs to the end, even if it's not going to work, ok? And so, countering this kind of incentive is, is, is very important to be successful, because you want to fail fast and…”
Disclosure
BenevolentAI cannot discover unstudied targets; it only repurposes existing knowledge
“We don't come up with new targets that have never been looked at. For example, the system by definition cannot do that. So it's always about repurposing some existing knowledge.”
Assertion Supported
Pesenti: Facebook uses embedding algorithms to automatically match misleading content
“And then when they fly content that should be at least, you know, shown as a misleading, then we have this very advanced similarity algorithm that kind of like look at an embedding of the of the content itself. Again, it can be the image itself or a multimodal…”
Assertion Not checkable as stated
Pesenti: Facebook gained a year of usage growth in one month
“Basically the growth that happened in, just a matter of a month was what we predicted would happen over more than a year”
Assertion Not checkable as stated
Pesenti: Alexa and Google Home rely on limited, scripted dialogues
“When you interact with something like Alexa or Google Home today, right, this system basically do intent recognition and then they have a very scripted dialogue after this, right? So usually You recognize one of the steps, and it's very limited because you can…”
