Cyril Labidi

Anthropic just released Claude 3.5, a new model that surpasses GPT-4o on multiple benchmarks while being 2x faster and 80% cheaper than Claude 3 Opus. The model is already available for free on http://claude.ai and the Claude iOS app. Claude 3.5 Sonnet is their our strongest vision model. Sonnet now surpasses Claude 3 Opus across all standard vision benchmarks. Improvements are most noticeable in tasks requiring visual reasoning, like interpreting charts, graphs, or transcribing text from imperfect images. https://lnkd.in/dbmeu4ms https://lnkd.in/d3UjW8Ee

14

1 commentaire

Gaussian Ray Tracing: Fast Tracing of Particle Scenes Nicolas Moenne-Loccoz, Ashkan Mirzaei, Or Perel, Riccardo de Lutio, Janick Martinez Esturo, Gavriel State, Sanja Fidler, Nicholas Sharp, Zan Gojcic NVDIA 2024 https://lnkd.in/d4ur_SBY https://lnkd.in/d-Qg-68Y - The paper proposes a GPU-accelerated ray tracing algorithm for rendering particle-based radiance field scene representations like 3D Gaussians. - To build a Bounding Volume Hierarchy (BVH), particles are encapsulated in stretched polyhedron meshes that tightly bound them. This allows leveraging fast ray-triangle intersections during tracing. - For efficiency, the rendering algorithm gathers and sorts batches of k closest particle hits using a k-buffer along each ray before evaluating their contribution. - The proposed ray tracer enables advanced graphics techniques like reflections, refractions, depth of field, as well as training with random rays. - Experiments show the ray tracer achieves real-time performance and quality on par with state-of-the-art rasterization methods, while also enabling new applications. - Generalized Gaussian kernels are proposed which reduce particle hits and improve performance. The sharper fall-off better encapsulates particle response. abs: Particle-based representations of radiance fields such as 3D Gaussian Splatting have found great success for reconstructing and re-rendering of complex scenes. Most existing methods render particles via rasterization, projecting them to screen space tiles for processing in a sorted order. This work instead considers ray tracing the particles, building a bounding volume hierarchy and casting a ray for each pixel using high-performance GPU ray tracing hardware. To efficiently handle large numbers of semi-transparent particles, we describe a specialized rendering algorithm which encapsulates particles with bounding meshes to leverage fast ray-triangle intersections, and shades batches of intersections in depth-order. The benefits of ray tracing are well-known in computer graphics: processing incoherent rays for secondary lighting effects such as shadows and reflections, rendering from highly-distorted cameras common in robotics, stochastically sampling rays, and more. With our renderer, this flexibility comes at little cost compared to rasterization. Experiments demonstrate the speed and accuracy of our approach, as well as several applications in computer graphics and vision. We further propose related improvements to the basic Gaussian representation, including a simple use of generalized kernel functions which significantly reduces particle hit counts. #AI #graphics #computervision #patternrecognition

23

Datasheets are the primary communication channel for the electronics and semiconductor industry. This blog post from Wizerr AI team outlines what makes a good datasheet - for all stakeholders.

24

// Perceptual Quality Measurement and Optimization of Displays // Perceptual models have been leveraged to optimize the quality of XR displays, including improved rendering speed, reduced data volumes, or reduced target-searching times. This award-winning paper examines this type of optimization in the context of saving display power.

1

🦙 Meta AI Releases Llama 3: 8B & 70B Models! 🚀 Meta has released two new Llama 3 models : 8B parameter model outperforms 70B Llama 2 🤯 70B parameter model is competitive with Gemini Pro 1.5 and Claude 3 Sonnet 💪 (more models are coming soon!) 405B model in the works, nearing GPT-4 level 👀 🔑 Key Takeaways: Trained on 15T tokens, 7x more than Llama 2 8K context window, shorter than some recent models Performs well on reasoning benchmarks like GSM8K Supports commercial use with some license restrictions 🧪 Trying it Out: Available via Ollama, HuggingFace, API services Can be run locally with quantization Responds well to chain-of-thought & ReACT prompts Some multilingual support but dedicated model likely coming 📜 License Notes: - Can't use Llama 3 to train other large models - Must include "Llama 3" in name of fine-tuned versions - Review prohibited use cases before commercial use The Llama 3 models look promising and signal continued rapid progress in open LLMs. Expect to see powerful fine-tuned versions and new capabilities soon! 🦙🔥 #Llama3 #OpenLLMs #LLMProgress

74

2 commentaires

This narrative is totally true. However, functions of Tokens in LLMs are: 1. Breaking down Text into manageable pieces. These tokens van be individual words, punctuation marks et cetera 2. LLMs can analyze language structure more effectively. It helps you to grasp relationship between words, grammatical forms and sentence narratives. 3. Context and Cost. For Context, it is great for Semantic search projects. Some of my products run on this Architecture. For Cost, I have to deal with this regularly. I manage AI products that are cost effective but potent to deliver on the Stakeholder's ask. You can learn a lot here by DeepLearning.AI: https://lnkd.in/esXqR5fj Signed: Temitope Oladokun - UNDYING RESILIENCE. #llm #token #ai #technology #technicalproductmanager #productmanagement

1

For those seeking investors and for those making investments - a great thread by Nicole DeTommaso with important advice: Choose your key points but not all of them are available. Read 👇👇👇 #vc #investing #founders #startups

2

Deep tech founders, as it has been said before. Yes, design is hard. But manufacturing your design is 10x harder. More often than not, I see revenue projections but no: 1 / Establishment to the first order the manufacturing process has been validated for operation at reasonable scale. 2 / What it will take to ramp up in a few years to the very high production levels indicated by the revenue projections. I am telling you, going through the heartache of figuring this out sooner than later will save you more painful heartache over a greater stretch of time. We all know that execution beats ideas. This is a major part of execution, if not nearly all. And it goes for beyond deep tech but anyone bringing anything physical to life for broad consumption. The magic of cars isn't necessarily that one was invented. The magic is that we were able to get one into nearly everyone's hands. Thoughts???

743

140 commentaires

Learn from OpenAI’s playbook: Prioritizing key projects and empowering a mission-oriented team are essential for startups aiming to make a significant impact 👇 ➡ 𝗜𝗻𝘁𝗲𝗿𝘃𝗶𝗲𝘄𝗲𝗲: Sam Altman, CEO at OpenAI and Brad Lightcap, COO at OpenAI 🔌 𝗜𝗻𝘁𝗲𝗿𝘃𝗶𝗲𝘄𝗲𝗿: Harry Stebbings, Founder at 20VC ✏ “20VC: OpenAI’s Sam Altman and Brad Lightcap on The Future of Foundation Models” ⭐ 𝗙𝗼𝗰𝘂𝘀 𝗼𝗻 𝗯𝘂𝗶𝗹𝗱𝗶𝗻𝗴 𝗮𝘀𝘀𝘂𝗺𝗶𝗻𝗴 𝘁𝗵𝗮𝘁 𝗔𝗜 𝗺𝗼𝗱𝗲𝗹𝘀 𝘄𝗶𝗹𝗹 𝗰𝗼𝗻𝘁𝗶𝗻𝘂𝗲 𝘁𝗼 𝗶𝗺𝗽𝗿𝗼𝘃𝗲 𝗮𝘁 𝗮 𝗿𝗮𝗽𝗶𝗱 𝗽𝗮𝗰𝗲: "There are two strategies to build on AI right now: [...] (1) to assume the model is not going to get better and then you kind of build all these little things on top of it; [...] (2) to build assuming that OpenAI is going to stay on the same rate of trajectory and the models are going to keep getting better at the same pace." ⭐ 𝗙𝗼𝗰𝘂𝘀 𝗼𝗻 𝗮 𝘀𝗺𝗮𝗹𝗹 𝗻𝘂𝗺𝗯𝗲𝗿 𝗼𝗳 𝗰𝗿𝗶𝘁𝗶𝗰𝗮𝗹 𝗽𝗿𝗶𝗼𝗿𝗶𝘁𝗶𝗲𝘀 𝗮𝘀 𝗮 𝗰𝗼𝗺𝗽𝗮𝗻𝘆 𝘀𝗰𝗮𝗹𝗲𝘀: "At any given point in a company's life, there's only one to three things that really matter. And that spills over into how we run the team because [...] if we can align on what those things are, [...] then it helps me to translate down to the teams that I'm building." ⭐ 𝗘𝗺𝗽𝗼𝘄𝗲𝗿 𝗲𝗺𝗽𝗹𝗼𝘆𝗲𝗲𝘀 𝘄𝗶𝘁𝗵 𝗮𝗰𝗰𝗲𝘀𝘀 𝘁𝗼 𝗔𝗜 𝘁𝗼𝗼𝗹𝘀 𝘁𝗼 𝗶𝗻𝗰𝗿𝗲𝗮𝘀𝗲 𝗽𝗿𝗼𝗱𝘂𝗰𝘁𝗶𝘃𝗶𝘁𝘆: "I think people criminally underrate how important it is and how much return you really get on just giving people access to the technology. [...] Someone who used to spend two days doing something now spends two minutes doing it and is freed up to do 85 other things in their daily life." ⭐ 𝗣𝗿𝗶𝗼𝗿𝗶𝘁𝗶𝘇𝗲 𝗺𝗶𝘀𝘀𝗶𝗼𝗻 𝗼𝗿𝗶𝗲𝗻𝘁𝗮𝘁𝗶𝗼𝗻 𝗮𝗻𝗱 𝗵𝗶𝗿𝗲 𝗽𝗲𝗼𝗽𝗹𝗲 𝘄𝗵𝗼 𝗮𝗿𝗲 𝗴𝗲𝗻𝘂𝗶𝗻𝗲𝗹𝘆 𝗲𝘅𝗰𝗶𝘁𝗲𝗱 𝗮𝗯𝗼𝘂𝘁 𝘁𝗵𝗲 𝗰𝗼𝗺𝗽𝗮𝗻𝘆'𝘀 𝗳𝘂𝘁𝘂𝗿𝗲 𝘃𝗶𝘀𝗶𝗼𝗻: "I watched what has happened to other tech companies when they just become the place you want to work because it's a good resume item. [...] I think companies that lose their mission orientation and get taken over by mercenaries usually come to regret that." ⭐ 𝗥𝗲𝗰𝗼𝗴𝗻𝗶𝘇𝗲 𝘁𝗵𝗮𝘁 𝘁𝗵𝗲 𝗯𝗲𝘀𝘁 𝗶𝗱𝗲𝗮𝘀 𝗼𝗳𝘁𝗲𝗻 𝗰𝗼𝗺𝗲 𝗳𝗿𝗼𝗺 𝘂𝗻𝗲𝘅𝗽𝗲𝗰𝘁𝗲𝗱 𝗽𝗹𝗮𝗰𝗲𝘀 𝘄𝗶𝘁𝗵𝗶𝗻 𝘁𝗵𝗲 𝘁𝗲𝗮𝗺, 𝗻𝗼𝘁 𝗷𝘂𝘀𝘁 𝘁𝗵𝗲 𝗺𝗼𝘀𝘁 𝗲𝘅𝗽𝗲𝗿𝗶𝗲𝗻𝗰𝗲𝗱 𝗵𝗶𝗿𝗲𝘀: "Find a way to make sure that there's this very flat and even playing field when it comes to how you look to the team for perspective, decision-making, judgment, and creativity." ───── 👉 Follow me, Alexander Small, for tactics and strategies for building startups from industry-leading Founders, Operators and Investors

4

Little can be added to those statements. Hardware can be one of your best competitive advantages, if done right.

5

1 commentaire

‘hype cycle’ Kevin Mayer says AI hype is at an all-time fever pitch right now : warned of a coming “stabilization” for valuations : AI artificial intelligence vastly overhyped already: Former TikTok CEO around artificial intelligence has reached a crescendo valuations looking “astronomical” “AI artificial intelligence provides capabilities that have not yet been seen and are very valuable the hype cycle has been dramatic” “I think we’ll see a peak of that, of the hype within AI artificial intelligence, the valuations, and everyone talking about how it’s gonna disrupt every single corner of our economic universe and personal lives” artificial intelligence is “vastly overhyped already” “Too many companies [are] bringing in too much capital at valuations that are way too astronomical So there will be a stabilization. There’ll be a realization of the benefits of AI artificial intelligence in many, many industries, but I think the hype is at an all-time fever pitch right now” ---- generative artificial intelligence ---

An elusive tech co-founder: the startup's Holy Grail 🏆. But do you deserve them? 🤔 Great tech co-founders have endless options, and they're not just looking for ideas💡. They want traction 🚀. So, how can you become attractive to top tech talent? Show your strengths 💪: 1️⃣ Pre-sell your product – generate initial interest 2️⃣ Build your personal brand – create a buzz around your project 3️⃣ Garner email subscriptions/orders – provide services manually if needed Proving you can get traction will make you more appealing to potential co-founders. It's not about funding; it's about results ✅. Instead of chasing that rare unicorn 🦄, focus on building your business, making connections, and proving your worth. When you're ready, the tech co-founder you've been dreaming of may just walk through the door 🚪. #startups #founders #entrepreneurship

7

6 commentaires

Sam Altman on talent investing (2019 YC era): “I have become so convinced that the only thing I have to do, that my firm has to do, is find the smartest, most talented people in the world, discover those people, and enable them. Make bets on them as people and trust that they will figure out, over time, they will drift towards the right ideas and the right kind of businesses. … One of the most fun things about YC is that, I think, we have more data points on what successful founders and bad founders look like than any other organization has had in the history of the world. We have that all in our heads, and that’s great. So I can say, with a high degree of confidence, that this correlation is true. Being a fast mover and being decisive — it is very hard to be successful and not have those traits as a founder. Why that is, I’m not perfectly clear on, but I think it is something . . . about the only advantage that startups have or the biggest advantage that startups have over large companies is agility, speed, willing to make nonconsensus, concentrated bets, incredible focus. That’s really how you get to beat a big company. .. Paul Graham used to say that the prototype for a good founder was James Bond. There’s lots of reasons the analogy doesn’t work, but this idea that a lot of things go wrong, you are unflappable, you never lose your cool, you know how to react. Bad things are happening to you left and right, and you are a survivor, and you always figure it out with whatever resource is in front of you. That part is the spirit of a great founder.” https://lnkd.in/gtB8xNAe

14

1 commentaire

Played around with GPT-4o. It's fast, powerful, fun and most importantly FREE. Why is it free? My take : • OpenAI wants to have a billion users before they think of money. Case in point - Last week, I sent a message in a founder group for testing Butternut AI GPT. 18/23 folks no longer had the paid plan! It is becoming a wider trend - people had started unsubscribing GPT Plus plan as free is good enough for most users. • GPT Store has amazing GPTs but only the paid subscribers had access to it. With the free access, the usage of GPTs will increase 100x or more. Once businesses make money, they'd happy to share revenue with OpenAI. • Freemium is a proven model in consumer companies. Once you've 100M+ users, there are so many ways to monetize. • Premium users and businesses can easily subsidise free users like they have done in case of Facebook, Google, LinkedIn and more. • OpenAI has an early mover + brand advantage. But with launch of Claude, Gemini and others - the competition is heating up. Giving free access will help OpenAI leverage on the brand name. All in all, making it free has open floodgates for AI usage and ChatGPT. Super excited to build in this everyday evolving AI space. 📷 Excerpt from Sam Altman's latest blog on GPT-4o

39

4 commentaires

How far will Tesla shares drop? They just published Q1 results. Not good, to say the least. On the other hand from user perspective Tesla Model 3, especially with subsidies, represents remarkable value for money! It is priced at €40,000 in Poland🇵🇱 (subsidised) Tesla Company market overview: Pros: - The Model 3 and Y still outperform competitors. Cons: - A 13% year-over-year revenue drop indicates potential market saturation or lost market share. - Net income tumbles 55% - Tesla’s engagement in a price war with China. Honestly? Price war vs China? 🇨🇳 - Intensifying competition from both budget and luxury segments, particularly from Chinese manufacturers. - Criticism over halo products and marketing strategies build up. - Uncertainty surrounds the future product lineup. - Customer feedback suggests dissatisfaction with some design choices, like indicators built into the steering wheel and gearbox integrated on the screen? - quite foggy AI, digital, robot outlook to be honest. Can Elon Musk refocus on core business and client needs? Can he just follow Mark Zuckerberg’s recent strategic shifts? #tesla #elon

5

10 commentaires

Focusing on vanity metrics is the fastest way to undermine growth. Many of the most visionary founders I've worked with mistake vanity metrics for product<>market fit. This is not breaking news, but I still see it all the time. Too many times I've seen visibility poorly translated for value. But that's not how the best do it. Instead, they focus on metrics that reflect *real* user value. Here are 7 steps that will help you build a sustainable growth engine: 1. Define your North Star Metric *** Hint: It's not revenue. It's one metric that defines your high-engaged users, the ones that would be upset if your product or service disappeared 2. Identify leading indicators that drive the NSM and waterfall those into department-level OKRs *** I like cascading OKRs into OKRAs at the team or individual level, with a bottoms-up on Actions 3. Ruthlessly prioritize efforts within your team to move those leading indicators *** This can be the hardest part for high-growth companies: Focus 4. Cut efforts that don't show material impact on monthly/quarterly indicators *** Like yard signs in politics, executive leadership loves to "see" the work being done (e.g. social media) but these rarely move the needle alone 5. Implement a weekly meeting structure that allows for bottoms-up autonomy but still provides time for feedback and unblocking 6. Fail fast, but learn quicker *** It's okay to experiment but if you eat what you kill, you better learn to hunt. 7. Double down on winning bets; rinse, repeat *** It's okay to experiment but if you eat what you kill, you better learn to hunt. Remember — you're not chasing how many people know about you, but how many people can't live without you. 📈

9

4 commentaires

The idea is easy. Manufacturing and mass production is the difficult piece of growing any product.

5

Open-source AI models such as Llama2, Llama 3, and Mistral are making strides toward ensuring AI technology isn't monopolized by a few companies. However, the primary barrier remains the hardware requirements and associated costs needed to run these models. Although it's not yet fully democratized—thanks to a continuing hardware monopoly by companies like NVIDIA, reminiscent of the old adage, "sell shovels during the gold rush"—AI is becoming increasingly accessible to most developers through AI platforms like https://huggingface.co/, https://openai.com/, https://www.anthropic.com/, https://groq.com/, ...

51

2 commentaires