![AI Master Group](/img/default-banner.jpg)
- 46
- 45 037
AI Master Group
Приєднався 16 жов 2023
Mesh Anything (except a Pink Hippo Ballerina)
The developers at MeshAnything have just released new code that offers an important improvement in how the surface of 3D objects can be encoded. What the new method does is build out the shape by always seeking to find and encode an adjacent face that shares an edge, which requires only about half as many tokens to represent the same information by other methods, resulting in a four-fold reduction in the memory requirement to achieve the same task, which enabled MeshAnything to double the maximum number of faces it can handle on a single object to 1600, as compared to 800 for current methods.
This video starts by comparing the new method with the current one. After that, we generate a 3D object from a text prompt on the Rodin website (a pink hippopotamus ballerina character with white tutu), and we check it on the Sketchfab website. Then we run the code that was provided by MeshAnything on GitHub, and we check the output on Sketchfab, comparing before and after side-by-side.
The results confirm the final words of the paper, which state that “the accuracy of MeshAnything V2 is still insufficient for industrial applications. More efforts are needed.” Nonetheless, this new computational approach is elegant, and the video concludes with a prediction that we’ll likely see improvements that build on the foundations laid by MeshAnything V2.
This video starts by comparing the new method with the current one. After that, we generate a 3D object from a text prompt on the Rodin website (a pink hippopotamus ballerina character with white tutu), and we check it on the Sketchfab website. Then we run the code that was provided by MeshAnything on GitHub, and we check the output on Sketchfab, comparing before and after side-by-side.
The results confirm the final words of the paper, which state that “the accuracy of MeshAnything V2 is still insufficient for industrial applications. More efforts are needed.” Nonetheless, this new computational approach is elegant, and the video concludes with a prediction that we’ll likely see improvements that build on the foundations laid by MeshAnything V2.
Переглядів: 1 204
Відео
Can Robots Win at Table Tennis? Take a Look!
Переглядів 2,1 тис.14 днів тому
Google DeepMind has just achieved a new level of robotic skill - the ability to compete and win at table tennis, a game that requires years of training for people who want compete at an expert level. This video shows the robot in action against an array of competitors, ranging from beginner level to tournament pro and, in doing so, describes both the hardware and AI aspect, including how it was...
Shark Alert! YOLO AI-Vision in Action
Переглядів 1,4 тис.21 день тому
Last week, several news outlets ran a story about SharkEye, which is an AI-vision shark detection program, developed at the University of California, Santa Barbara, and deployed at California’s Padaro Beach, which is an area where surfers and great white sharks are both frequently found. After quickly describing the program itself, the video identifies the underlying technology that was used fo...
AI Can do That?? Silver Medal in Pure Math
Переглядів 1,7 тис.Місяць тому
AI has just achieved an amazing milestone. A couple of Alpha models by Google DeepMind scored silver-medal-level performance in a globally-recognized competition in advanced mathematics: IMO 2004. This video starts by setting the context for this latest achievement, going back to significant milestones in 2022 and 2023 that helped set the stage for what just happened, sharing the story along th...
Will Open-Source Llama Beat GPT-4o?
Переглядів 618Місяць тому
Last week Meta launched its newest family of models, Llama 3.1, including a new benchmark - an open-source foundation model with 405 billion parameters. With this, Zuckerberg predicted that Meta AI will surpass OpenAI’s 200 million monthly active users by the end of this year. Hubris aside, this video looks at six reasons why we need to pay attention to this announcement, including Zuckerberg’s...
Call a Doctor! --Blue Screen Lessons Learned
Переглядів 983Місяць тому
Companies worldwide grappled on Friday with what Troy Hunt, famously described as “the largest IT outage in history,” caused by a faulty sensor configuration update that got pushed to Microsoft by the cyber-security giant, CrowdStrike, resulting in a $31 billion loss in market capitalization for the company. Specific information about the bug is not yet publicly available, but this video presen...
Amazing Milestone! Million Experts Model
Переглядів 1 тис.Місяць тому
A top researcher at Google DeepMind just released an important paper, “Mixture of a Million Experts.” As the paper’s title announces, it describes an approach that resulted in the first-known Transformer model with more than a million experts. For context, the number of experts currently seen in smaller models varies between 4 and 32, and ranges up to 128 for most of the bigger ones. This video...
Behind the Curtain of Figma AI
Переглядів 584Місяць тому
The recent announcement of Figma AI generated both excitement and controversy. This video summarizes the new AI features in under three minutes, for this popular design tool that’s used for creating prototypes of digital experiences. Next, the video looks at the underlying technology that was used to enable the new AI features, including OpenAI language models and the Amazon Titan diffusion mod...
How a Language Model Aced a Top Leaderboard
Переглядів 1,2 тис.2 місяці тому
This video shares details about a remarkable experiment by researchers in Tokyo, who teamed up with Oxford and Cambridge Universities to study whether large language models might now be able to write code that improves their own performance. The answer was Yes. Not only that, the model created a whole new approach that placed it at the top of a leaderboard, using a novel method that had not yet...
New Method Runs Big LLMs on Smartphones
Переглядів 1,9 тис.2 місяці тому
There’s a big breakthrough that just came out for handling large language models on smartphones. It’s called PowerInfer-2 and what it does is look at every option for a processing an LLM on a particular smartphone, and picks the fastest way for that particular LLM on that particular device. For example, it uses completely different computation patterns for the early vs. the later phases of the ...
Nemotron-4 is BIG in More Ways than One
Переглядів 7722 місяці тому
Last week, NVIDIA announced Nemotron-4, which consists of three models: Base, Instruct and Reward. These three models work together within the NeMo framework to enable the creation and fine-tuning of new large language models. At 340 billion parameters, this new entrant far bigger than any other open source model, but the really big news is that Nemotron-4 comes with a permissive license that a...
Testing Ollama on Hard Questions
Переглядів 1,2 тис.2 місяці тому
Ollama is a popular platform for running language models on your local machine, with access to almost 100 different open source models, including llama-3 from Meta, Phi3 from Microsoft, Aya 23 from Cohere, the Gemma models from DeepMind and Mistral. This video shows llama-3 being run on a laptop, using Ollama. Three difficult questions are presented in turn to each of GPT-4o, Gemini and llama-3...
Hacking Passwords with ChatGPT?
Переглядів 1,5 тис.3 місяці тому
The latest edition of the Hive Systems password table is now available, and it shows ChatGPT as the fastest option by far, for hacking passwords, which certainly requires some explanation! This video looks at the assumptions that go into time is takes for a hacker to get a password by brute force. Along the way, we look at hashing algorithms like MD5 and bcrypt, and we look at hardware like NVI...
What is AGI? --the Ultimate Test!
Переглядів 7213 місяці тому
Since there’s lots of attention right now on AGI, it’s time to finally define what that is - digging deeper into the underlying implications of these three words: “artificial general intelligence,” and producing in a succinct one-sentence definition. This video reviews information suggesting that we either have AGI already now, or we are very close to having that. Along the way, we distinguish ...
GPT-4o Rapid Fire Highlights
Переглядів 1,6 тис.3 місяці тому
The launch of GPT-4o is a big deal. Here's a rapid-fire summary of the highlights. This video is a mix down of the 5 key announcements from the original 26 minute video in under one minute. Then, you get a rapid-fire demo of 7 key abilities of GPT-4o in under 7 minutes. You will certainly be amazed. By the way, does that voice sound like it’s from Scarlett Johansson? You be the judge. . .
Virtual AI Announcers - Good Better Best
Переглядів 1,8 тис.5 місяців тому
Virtual AI Announcers - Good Better Best
How to Make (Even More) Money with Generative AI
Переглядів 8795 місяців тому
How to Make (Even More) Money with Generative AI
Enter the "Chief AI Officer"! (... what's that?)
Переглядів 1 тис.5 місяців тому
Enter the "Chief AI Officer"! (... what's that?)
Sora Preview: OpenAI's Text to Video Surprise!
Переглядів 4,2 тис.6 місяців тому
Sora Preview: OpenAI's Text to Video Surprise!
Has AI Learned to Lie? New Findings!
Переглядів 8206 місяців тому
Has AI Learned to Lie? New Findings!
Future of AI? New Network Solves Problems Differently
Переглядів 1,3 тис.6 місяців тому
Future of AI? New Network Solves Problems Differently
Copilot Showdown? Azure vs AWS vs GCP
Переглядів 9136 місяців тому
Copilot Showdown? Azure vs AWS vs GCP
Stable Code 3B: Hype or Hero or Cats with Hats?
Переглядів 1,7 тис.7 місяців тому
Stable Code 3B: Hype or Hero or Cats with Hats?
I'm not even a particularly advanced player and I guarantee I would destroy that robot. It looks to be barely beyond advanced amateur in my opinion. Any club player would outscore the robot 2 to 1 or more. Most club players smash anything that goes over the net by more than about 6 inches (and often a lot less than that). To advance very much it's speed needs to increase by a factor of 2 or 3.
Here’s a link to the code I ran in the video. github.com/buaacyw/MeshAnythingV2 And here’s a link to the Rodin website where I created the 3D character. hyperhuman.deemos.com/rodin
Thank you ! It's a really funny video, and I love your calm voice ^^
"he" this, "him" that. Is that thumbnail narrator AI-generated?
That's me: Jim Griffin. I'm a real person, not AI generated. I hope the question implies that I sounded professional!!
Could not deal with the massive spin of advanced players.
Maybe But once it gets the data ... The robot will strike very early So it'a all about the input.
@@thereistheonlyone Yes indeed, if it was able to receive the input and calculate position AND movement needed at point of contact. Because it would also need movement options since just getting its racquet in the right position (with slight movement) is not enough to deal with some spins. We have to provide counter-spin at times, which is not a simple matter to determine or execute.
@@thereistheonlyone And on rewatching some, I didn't even see it deal with basic back spins.
Here’s a link to the full paper: Achieving Human Level Competitive Robot Table Tennis arxiv.org/pdf/2408.03906
Here’s a link to the SharkEye website. sharkeye.org/#our-process And here’s a link to the documentation page for YOLO. docs.ultralytics.com/
they should use this to end world hunger and do cancer research pleaseeeeeeeeeeeeeeeeeeeeeeee
Here’s a link to the full article from Google DeepMind. AI achieves silver-medal standard solving International Mathematical Olympiad problems deepmind.google/discover/blog/ai-solves-imo-problems-at-silver-medal-level/
Here’s a link to the blog post I quoted from in the video. Open Source AI Is the Path Forward about.fb.com/news/2024/07/open-source-ai-is-the-path-forward/
If you’re interested in digger deeper into this topic, I highly recommend this excellent article by Ed Bott on ZDnet, which includes some very helpful historical context. “What caused the great CrowdStrike-Windows meltdown of 2024? History has the answer.” www.zdnet.com/article/what-caused-the-great-crowdstrike-windows-meltdown-of-2024-history-has-the-answer/
Great video! Thank you for the clear explanation. The calm tone made it easier to undertand.
That's very nice of you. Thank you very much!
Here’s a link to the paper I featured in this video: “Mixture of A Million Experts,” by Xu He at Google DeepMind arxiv.org/pdf/2407.04153
At the very end of this video there’s a snippet from a brilliant performance of the song “Daisy Bell: A Bicycle Built For Two,” by permission from Julien Neel. (Thank you Julien!) Here’s his website: julienneel.sellfy.store/ All four singers in the barbershop quartet are Julien. It’s brilliant - and amazing. Here’s a link to the full song on UA-cam. ua-cam.com/video/JmiNlZAiDqo/v-deo.html
Here’s a link to the paper this video is about: “Discovering Preference Optimization Algorithms with and for Large Language Models” arxiv.org/pdf/2406.08414
Please check your microphone volume before upload, you are DRAMATICALLY quieter than any of the sound effects or your end card. I'd hazard a guess at around 12dB quieter.
Thanks for the heads-up. I need to study up on how to Normalize volume using Essential Sound in Adobe Premier Pro. (I know how to do that in Sound Forge.)
You could probably get it running at home through a smart scalar API like Ollama with 4 A6000s, 512gb of system RAM, and a pair of Intel Scalable CPUs with AVX-512 cores, it'll be around $10k to build on used parts. It won't be snappy, 2 minutes per response at least, but it's cheaper than renting a whole DGX for a month. A 22.3k entry synthetic dataset generated by continually inferring desireable results with Nemotron over a month should be good enough to start training a smaller model to a far higher standard than it's natural dataset counterpart could provide, and look at that, you'll also have hardware good enough for training 32B models already on-hand by the end.
That's an interesting idea. It's worth a try.
That's mixtral though isn't it just getting memory throttled
Yes, you're right. Most devices would certainly get memory throttled running Mixtral-47B. PowerInfer-2 apparently helps to mitigate that issue, both by optimizing the computation approach and by selecting processing units in a more planful way.
No performance gains on CPU unless you use their models *and* you have avx2 instruct set, so meh. I'll stick to my current project of making ultra slim, CPU only mini models that challenge mainstream, thanks.
Sounds like an interesting area of focus: Ultra-slim, CPU-only mini models. You're right: PowerInfer-2 optimizes for various processing units, including CPU configurations that support AVX2 instructions. Although you're constrained on the CPU, you still might get a performance improvement from other aspects of PowerInfer-2 -- especially the approach they describe to handling sparse data, so it still might be worth exploring . . .
Very interesting. Is there a way to link powerinfer2 to sillytavern ?
don't make such loud noises in the video
phone farms might take this on 🍿
Here’s a link to the paper I cited in the video. PowerInfer-2: Fast Large Language Model Inference on a Smartphone arxiv.org/abs/2406.06282
Agent smith.
As promised in the video, here’s a link to the technical report we looked at. Nemotron-4 340B Technical Report d1qx31qr3h6wln.cloudfront.net/publications/Nemotron_4_340B_8T_0.pdf
I like this test! However I think the first math question was very “textbook” in nature such that if it understood the key words could probably directly copy from source text. Similar to canterberry tales. But the middle question required specialized knowledge and further subtle understanding. Nice to know gtp4.0 isn’t perfect at least! At leas lt not yet…
I agree with you. I think the first question was kind of easy / textbook compared to the other two.
Here’s a link to the full list of models accessible via Ollama. There are 92 models currently listed there. www.ollama.com/library And here’s the download page I showed in the video. www.ollama.com/
My man Beautifully said
As promised in the video, here’s a link to the announcement about the new AI Master Group podcast, which launches on July 7. aimast.org/jim-griffin/
Here’s a link to the paper I cited by Bowen Xu, “What is Meant by AGI? On the Definition of Artificial General Intelligence” arxiv.org/html/2404.10731v1
Here’s a link to the announcement page, which has quite a few additional videos and images beyond what I showed. Hello GPT-4o : openai.com/index/hello-gpt-4o/
Hope you are having a wonderful weekend, Jim.
Thank you to Lemon for permission to use parts of the track, Mr. Bongo! Here's a link to Lemon / Freshly Squeezed on Sound Cloud soundcloud.com/freshlysqueezed and here's a link to Ursula1000 soundcloud.com/ursula1000
As promised in the video, here's a link to the original paper with technical details about BASE TTS. assets.amazon.science/6e/82/1d037a4243c9a6cf4169895482d5/base-tts-lessons-from-building-a-billion-parameter-text-to-speech-model-on-100k-hours-of-data.pdf Best wishes!
Nice summary. But the engine used for Mona Lisa /Audrey Hepburn seems fundamentally different from the "AI announcers" because it obviously requires an input video of someone talking, which it applies to the photo - rather than generating new facial movement solely from text like the others. So the Mona Lisa one seems to have more in common with a deep fake.
nope. check the link.
@@mpcref I checked and EMO uses "input audio", so it's not generating the voice. Does seem to be generating the facial expression, but this isn't like a completely AI character as it uses your own audio.
@@Scripture-Man well yeah, no one claimed otherwise. The only input required is audio, not video. Whether that audio is AI generated or not doesn't make a difference.
Let me know if you want to connect to Ramsu or any of the founders at solus.ai They’re great people that I know well.
Audrey H dubbed Natalie Portman all in
As promised in the video, here’s a link to the page where you can find more information about EMO, including more amazing videos created by that tool. Check out “AI Girl generated by ChilloutMix,” and also Leonardo Wilhelm DiCaprio rapping Godzilla. EMO: Emote Portrait Alive - Generating Expressive Portrait Videos with Audio2Video Diffusion Model under Weak Conditions humanaigc.github.io/emote-portrait-alive/
Here’s a link to the thread I mentioned in the video from the OpenAI Developer Community page, where more than a dozen developers posted more than 60 different messages over a six month period of time, regarding pricing a Gen AI solution. There’s much more detail there than I could cover in a short video, so it’s worth a look. How to accurately price a gpt-4 chatbot? community.openai.com/t/how-to-accurately-price-a-gpt-4-chatbot/347250
As promised in the video, here’s a link to the tutorial by Patrick at AssemblyAI showing how to get started with Gradio. At about 8 minutes in, he shows an LLM being pulled into Gradio from Hugging Face. ua-cam.com/video/eE7CamOE-PA/v-deo.html And here’s a link to the blog I mentioned, written by Vellum founder Akash Sharma, showing the mechanics of testing competing LLMs against some of the metrics I mentioned in the video. www.activeloop.ai/resources/how-to-compare-large-language-models-gpt-4-3-5-vs-anthropic-claude-vs-cohere/
As promised in the video, here’s a link to the CNBC page with the story about Jeff McMillan’s promotion last week, including an embedded video, which is a previously-recorded interview between McMillan and CNBC’s Hugh Son about the recent generative AI project at Morgan Stanley. Morgan Stanley names a head of artificial intelligence as Wall Street leans into AI www.cnbc.com/2024/03/14/morgan-stanley-names-head-of-artificial-intelligence-jeff-mcmillan.html
As promised in the video, here’s a link to the video walk of FAISS through by James Briggs. James Briggs, Faiss - Introduction to Similarity Search ua-cam.com/video/sKyvsdEv6rk/v-deo.html And here’s a link to the page I showed, comparing 15 different vector databases. The link name says 2023, but it was updated again in 2024. Best 15 Vector Databases for 2024 [Top Picks] lakefs.io/blog/12-vector-databases-2023/
❤❤❤❤❤
People without souls 🫠😖😖😖👎👎👎
As promised, here’s a link to the Sora announcement page OpenAI: Creating video from text openai.com/sora And here’s a link to the technical report. Video generation models as world simulators openai.com/research/video-generation-models-as-world-simulators
Thx heaven for sphere packing
My uneducated opinion( you didn’t ask). Spin networking is your best bet to really understanding a space age. The rest is an early age of development wow
There are millions of problems to solve every micro second. Some math choices are more correct. Math is math
Does AI know how to encourage human desires to study. It seems to be addicting Without offering itself as a bonafide educator
"Redacted" on UA-cam looked at one AI company which produces 'historical images' and produced a Chinese female Nazi and a black male Nazi when asked to produce a profile of a Nazi. When asked to produce a picture of Elon Musk, Elon had turned black..! On Al-Jazeera, a couple days ago (it's the 2nd of March, 2024 today), AI was implicated in the Israeli bombings of civilians, women and children, in Palestine...
As I understood it there are no "reasoning abilities" which emerge. Its a feature to sound like coherent sentences made by reasoning but its a heuristic output based on its large trainingbase and some other factors. So in any cases it had enough trainingdata where similar events took place and it reacted to that pattern it doesn't mean that it can reason the same way in altered or more compicated scenarios. The "2 Jug " riddle example shows quickly how this apporach creates more unintended limitations sometimes.