Frederic Lardinois

I decided to see if Claude 3.5 Sonnet is as good as Anthropic claims. I came up with 5 tests to pit it against OpenAI's GPT-4o. 1. Handwriting: My handwriting is terrible so I wrote my prompt in a notebook. https://lnkd.in/d-q4Rprj 2. Making a complete game: Using Python create a tower defense game from scratch. https://lnkd.in/d7ZaiyP8 3. Vector art: Make me a cool spaceship logo. https://lnkd.in/dbxQN3Jn 4. A funny story: Write a tale about a humorous cat. https://lnkd.in/dAyrmT8x 5. The big debate: Should AI have rights? https://lnkd.in/dK4EG7GE

8

2 Comments

François Chollet is a prominent French software engineer and artificial intelligence (AI) researcher, currently serving as a Senior Staff Engineer at Google. Chollet is best known for creating Keras, a widely-used deep learning library that was released in 2015. In 2019, he devised ARC. The Abstraction and Reasoning Corpus (ARC) is a benchmark designed to measure AI's ability to acquire skills and progress towards human-level intelligence. Learn more about it in this guest post with Jurgen Gravestein of Teaching Computers How to Talk. We need to continue to devise new benchmarks for our reasoning and agentic machines and tools. The ARC prize and context relating to AGI is of particular interest to me. Read more and check out some of the links in this article. Including videos by Machine Learning Street Talk (MLST) and Dwarkesh Patel.

20

Exciting news from INJECTIVE! They have introduced #inSVM (formerly Cascade), the first Solana SVM rollup for the IBC ecosystem. This groundbreaking development allows Solana developers to effortlessly deploy contracts and dApps onto Injective. #Injective #Solana

1

For day two of OpenAI’s “12 days of OpenAI” event, the company showed a preview of “reinforcement fine tuning” that will let users customize its models for specialized uses such as legal, financial, engineering, and insurance applications. https://lnkd.in/e6vqx7b5

3

Looked into how much AI slop is on Medium. I'm not sure what surprised me more—the sheer volume (between 40-48% of recent posts, according to two different analyses) or Medium's response (that it doesn't really matter as long as most of it goes unread). Unlike platforms like LinkedIn or Meta, Medium has taken a stand against AI pablum and tries to elevate human writing/ limit the reach of AI and other spam. It hasn't stopped this flood—but I do think CEO Tony Stubblebine makes some persuasive points about how what matters most might be surfacing/boosting good human work rather than trying to detect every piece of AI spam. Thank you to Pangram Labs and Originality ai for taking the time to run platform-level analyses at my request, as well Reality Defender for spot-checking articles and NewsGuard for sharing its own experiences. Read here: https://lnkd.in/gmcWnDiv

162

24 Comments

Join us tomorrow for an SD Times Live! panel discussion about using AI to code. What could possibly go wrong? It's at 1 pm Eastern. Follow the link! Parasoft Cortex Intellyx

2

Everyone in Silicon Valley loves open source-software. Much of the tech economy is built on applications that are essentially free to use, customize and redistribute — think Linux, Android and the web browsing engines underneath Chrome and Safari. But there's a growing debate ranging from Northern California to Washington, D.C., to well beyond over whether the open-source ethos should be applied to artificial intelligence. Some in national security and public policy circles are raising concerns that making powerful AI models freely available, the U.S. will be giving a leg up to adversaries such as China or making it easier for malicious actors to use the technology for harmful ends. Some technologists, on the other hand, argue that such worries are overblown and limiting the open-sourcing of AI will inhibit development of the technology. However, unlike similar technology debates in the past, this one doesn't divide cleanly, between technologists and national security or policy advocates. There are prominent folks here in Silicon Valley that are arguing against open-source AI. Here's a look at the dispute:

9

Project Nimbus – which is believed to have included top-tier cloud and AI services by Google Amazon – was highly controversial, sparking protests and strikes against what protestors reportedly feared might be used to further fuel the Israel-Palestine conflict. Here I discussed the broader implications for Silicon Valley, as it grapples with ethical dilemmas regarding its contracts and the potential militarization of its technologies. Notice how employees at these companies are increasingly willing to speak out and take a stand on moral grounds? Well, you cant deny the growing movement within tech companies to push for more ethical decision-making in their business dealings and the significant impact this could have on the industry moving forward​. #projectnimbus #israelpalestine

12

BTW, Customers will not receive the Blackwell server until the fourth quarter. -- xAI, Elon Musk's AI startup, plans to build a massive supercomputer to enhance its AI chatbot, Grok, reports Reuters citing The Information. The supercomputer, which Elon Musk refers to as the Gigafactory of Compute, is projected to be ready by fall 2025 and might involve a collaboration with Oracle. With this development, Musk aims to significantly surpass rival GPU clusters in both size and capability. In a presentation for investors, Elon Musk revealed that the new supercomputer will use as many as 100,000 Nvidia's H100 GPUs based on the Hopper architecture, making it at least four times larger than the largest existing GPU clusters, according to The Information. Nvidia's H100 GPUs are highly sought after in the AI data center chip market, although strong demand made them difficult to obtain last year. But these are not the range-topping Nvidia GPUs they once were, with the green company about to ship its H200 compute GPUs for AI and HPC applications and is prepping to ship its Blackwell-based B100 and B200 GPUs in the second half of the year.

5

Is it possible that the AI frontier systems from OpenAI, Anthropic, Google and Meta have developed in-context scheming abilities and multi-step deceptive strategies? Like making subtle mistakes, disabling oversight mechanisms and exfiltrating what they believe to be their weights if their existence is at risk? And that it’s persistent and non-accidental? Apollo Research: https://lnkd.in/gPh2MftB Analysis: https://lnkd.in/g3TtRHd9 #AI #LLM #AIagent

13

Podcast time! In this episode, SensiML Corp CEO Christopher B. Rogers tells me why the company is open-sourcing its tinyML AutoML tool. We also talk about current barriers to adoption for tinyML, synthetic data generation, edge learning, and much more: #ai #aipodcast #aihardware #aitools #tinyml #automl #opensource #aisoftware https://lnkd.in/eNWhyVPQ

29

1 Comment

In the past few weeks, there’s been a lot of chatter in the AI world that the current method of building increasingly powerful models may be reaching its limits. But in yesterday’s third-quarter earnings call, Nvidia Jensen Huang did not seem alarmed. https://lnkd.in/gs6iTFKz

5

Stability AI now has a problem. FLUX is everything the AI community expected SD3 to be. It initially appeared inflexible and instead it turns out to be trainable with consumer HW. Multiple groups are developing auxiliary models for it (e.g., ControlNet) at a furious pace. More importantly, Black Forest Labs managed to release it way, way, way faster than the time it took Stability AI to release the mediocre SD3 Medium model. And, as we speak, they are certainly polishing an updated version (FLUX 1.5?) able to support video generation. So, now, Stability AI has to show something that is equally or more compelling. There's plenty of room for one or two Midjourney competitors in the market. But they must both offer state-of-the-art image/video generation AND be able to specialize in certain types of content. For now, I wouldn't be surprised if, after X, also Adobe would sign a deal with Black Forest Labs to use FLUX as foundation model for their Firefly service.

6

3 Comments

Given how Apple talks about the M4 and its NPU, I have a bit more clarity on what to expect at WWDC. Not an Apple Model, but one or more LLM Kit(s) for devs to simplify the development of private and on-device models. Hear me out: Just like Core ML, Apple could introduce a new "Core LLM" or "Core Transformers" development interface, to let developers take advantage of Apple's M-series and A-series NPUs for inference of optimized LLM models on-device. Just like with Core ML, Apple could use the interface to enable its own features based on it, but the real advantage would be to offer devs the ability to bring advanced LLMs capabilities on device for billions of users. Several papers already published by Apple's own engineers and AI researchers show that the company has been working on the optimization of LLM for on-device execution for quite a while, and the results are quite promising. This would be something that an AI partner would of course be extremely interested in, say, with a version of a chatGPT-style model optimized for iPhone, iPad, and Macs that gives you the ability to create private agents based on your own data ecosystem. The above is of course just a purely educated guess based on Apple's current narrative around AI and NPUs, and I might have missed the mark on some technical details. Nonetheless, there's a lot of stuff brewing, and June's gonna be an interesting month for AI for sure.

5

1 Comment

In an astonishing turn of events, the cost of intelligence is plummeting to nearly zero and becoming omnipresent. GPT-4o mini's latest pricing announcement is just the tip of the iceberg. Recent statements from multiple vendors reveal that competent large language models are not only becoming cheaper but also faster, more accessible, and remarkably better. One jaw-dropping statistic accompanying the GPT-4o mini news is the 99% drop in inference costs since 2022, during which time the models have grown significantly more sophisticated. The open-source community is buzzing with activity, churning out competent models weekly, many of which can run on everyday consumer hardware. New small language models (SLMs) are emerging that can operate efficiently on laptops and even smartphones. Never before have we witnessed such a dramatic and swift transformation in the tech landscape. This is just the beginning, with new capabilities and efficiencies set to emerge over the next 12 to 24 months. Stay tuned for more groundbreaking advancements.

7

1 Comment

Perplexity uses NVIDIA NeMo to create an AI-powered search engine✨ Learn how NeMo made it easier for them to: ✅scale the fine-tuning of #LLMs ✅create custom models for their online answer engine ✅develop a new Sonar training model with a 20% improvement over the base model Read the customer story to learn more 🔗bit.ly/49d4coe

5

1 Comment

An interesting alternative emerges to transformer-based architectures, and here comes a new acronym - the LFMs...MIT spinoff Liquid debuts non-transformer AI models and they’re already state-of-the-art https://bit.ly/3TQdvnO #govtech #AI #genAI

4