Nicolas Collignon

Are we as a species, mature enough, to create/develop a Benevolent-Beneficial #AGI/#ASI? I truly hope more people will realise how important and existentially crucial is for us to carefully govern the #AI evolution, and if we prove capable, when AI evolves into #BeneficialASI, in return this Benevolent/Beneficial #SuperIntelligence will guide/nurture humanity towards reaching our true/full potential as a species. Who knows, our potential is potentially limitless if we get this right. Progress often is disruptive/uncomfortable and necessary for a species' development/advancement/growth, but also in this mix, we need to add a pinch of caution and care too. It's a balancing act, a form of art in my opinion. It seems (we) are a species with an amnesia. We easily forget/ignore that we were lucky thus far to avoid extinction, and only after a cataclysm, and only then does humanity unite, and a period of growth/prosperity/innovation follows. Then after a few decades, the memory of the struggle fades and lessons learnt are gradually forgotten. Then something bad happens again and the cycle repeats itself. Well with the evolution of AI, being lucky won't be enough. Previous Cataclysmic events didn't possess intelligence greater than ours. Being strategically smart and cautious is more important than ever. All available fingers crossed that we still have a chance to create/achieve Benevolent #BeneficialAGI/ASI before the alternative, given the fact that misguided entities addicted/dependent on the hunger for power/dominance/control are better equipped/financed to accidentally create the alternative 😔. I ask myself, even if we are all united and work towards Beneficial #BenevolentASI together, Can we successfully do it without teaching it morality, while we still don't really/truly/deeply understand what morality really is? The lives of billions alive today and countless yet to be born are in the hands of the AI community. I truly hope we get this right, because if we do, the future will be better than we can imagine. You can find more about my scribbles on The Importance of The Responsible Development of Beneficial AGI here - https://lnkd.in/gUCiFaZq

11

5 Comments

#LLM features are no longer a mystery - they are being decoded. The recent paper from #Anthropic attempts to "map" the mind of their #LLM #Claude3. We will be able better understand the way models "think". Also, these features can be used to influence the model behavior to the desired one. A broad range of safety-oriented features are observed: including deception, sycophancy, bias, and dangerous content. Read more in the paper below: https://lnkd.in/e-s8rkyc #machinelearning

31

Here are the key points from the paper "Scaling Monosemanticity: Extracting Interpretable Features from Claude 3 Sonnet" by Anthropic: - Anthropic has scaled their sparse autoencoder (SAE) method to extract interpretable features from the large language model Claude 3 #sonnet. This allows them to decompose the model's activations into a sparse set of "feature directions" that represent abstract concepts. - The SAE consists of an encoder that maps model activations to a high-dimensional feature space, and a decoder that reconstructs the original activations from the feature activations. Training minimizes reconstruction error and an L1 penalty to encourage sparsity. - Anthropic found a diversity of highly abstract features in Claude 3 Sonnet, including features for famous people, countries, cities, and code type signatures. Many features are multilingual, and multimodal, and encompass both abstract and concrete instantiations of concepts. - The optimal number of features and training steps for the SAE scales as a power law with the compute budget used to train it. Larger SAEs achieve lower reconstruction losses. - Analyzing the learned features reveals that they become less specific as their activation strength decreases, possibly reflecting the model's confidence in a concept being present. - Extracting interpretable features from large language models like Claude 3 Sonnet could make them safer by revealing how specific features relate to concepts and affect the model's behavior. In summary, Anthropic has made significant progress in scaling its interpretability techniques to #llms, extracting millions of interpretable features that provide insight into the models' inner workings and representations. Source: https://lnkd.in/gqCsGR-Z

4

I took Meta’s brilliant paper, and asked Anthropic’s Claude 3.5 Sonnet to write a provocative article summarising it from the perspective of what this many mean for closed foundation models. Read below: Although, safety and alignment wise … it’s pretty easy to finetune these things if you know what you’re doing to turn them into subversive comedic monsters 😉🤔😂 Not that I would do such a thing. No sir! (full paper @ https://lnkd.in/d2qs6DPn ) —— Your Move, Sam Altman: Meta's Llama 3.1 Throws Down the Gauntlet In a bold move that has sent shockwaves through the AI community, Meta has unveiled Llama 3.1, a family of LLMs that not only rival the performance of OpenAI's vaunted GPT-4 but do so with a twist: they're giving it away for free. That's right Sam, the game has changed. While OpenAI has been guarding its golden goose behind an API paywall, Meta has decided to release not 1, not 2, but 3 powerhouse models into the wild. We're talking 8B, 70B, and a whopping 405B parameters of open-source AI goodness. It's like they're handing out Lamborghinis at a go-kart race. But it's not just about size (though let's face it, 405 billion parameters is nothing to sneeze at). Llama 3.1 is flexing on multiple fronts: Performance: The 405B model is going toe-to-toe with GPT-4 and Claude 3.5 Sonnet across a smorgasbord of tasks. It's like watching David take on Goliath, except David is also a giant and he's livestreaming the whole thing. Efficiency: While OpenAI's models are busy sipping mai tais in their cushy cloud servers, Llama 3.1’s 8B model can run on a toaster. Okay, maybe not a toaster, but certainly on your average laptop. Democratizing AI? More like starting an AI revolution. Transparency: Meta's not just showing their cards; they're explaining the whole trick. Training process? Here you go. Data curation? Have at it. Evaluation methods? It's all yours. It's so open, it makes OpenAI look like Fort Knox. Safety: They've got Llama Guard watching the gates. It's like they've hired a bouncer for their AI, but instead of keeping people out, it's keeping the AI in check. Multimodal Potential: Vision and speech capabilities are waiting in the wings. It's like Meta's saying, "Oh, you thought language was impressive? Hold my beer." So, what's the catch? There isn't one. Meta's basically dropped a nuke of openness on the AI landscape, and the fallout is going to reshape everything. To Sam Altman and the rest of the closed-source crew, the message is clear: the era of AI gatekeeping is over. Meta's not just inviting everyone to the party; they're giving away the DJ booth, the dance floor, and the whole damn club. In conclusion, Llama 3.1 isn't just a model; it's a manifesto. It's Meta saying, "We're not just in the game; we're changing it." So, Sam, the ball's in your court. Are you going to open up, or are you going to be left behind in a world where state-of-the-art AI is as open as a 24-hour diner? Your move, Sam. Your move.

7

1 Comment

Delighted to have contributed to the State of Wildfires 2023–2024 report by designing and developing two web applications that allow you to interactively explore the data that underpins this review article 🗺 📈 ➡ https://lnkd.in/eb-KY7Jy #wildfires #climatechange #dataviz #javascript

10

2 Comments

Very good explanation of lack of OOD generalization in LLM by Chollet. LLM is like the not so smart studious kid. “The other way you can define reasoning is as the ability to *synthesize* new programs (from existing parts) in order to solve tasks you've never seen before. Like, solving ax b=c without having ever learned to do it, while only knowing about addition, subtraction, multiplication and division. That's how you can adapt to novelty. LLMs *cannot* do this, at least not on their own. They can however be incorporated into a program search process capable of this kind of reasoning. This second definition is by far the more valuable form of reasoning. This is the difference between the smart kids in the back of the class that aren't paying attention but ace tests by improvisation, and the studious kids that spend their time doing homework and get medium-good grades, but are actually complete idiots that can't deviate one bit from what they've memorized. Which one would you hire?” #ai #llm #ood #generalization #reasoning #chollet https://lnkd.in/gVDX88dm

32

12 Comments

OpenAI O1 is creating a new wave of models that would be Thinking more before creating and "talking". Open Ai Is so confident about it that instead of naming the successor as GPT 5.xx the have reset the version to 1. Rakesh Gohel shares more insights into the O1 models

7