- Ben's Bites
- Posts
- AI chips are running out
AI chips are running out
PLUS: LLMs outside the US, new ways to use GPT
Hey folks, today we’ve got state of AI chips that run these models: Nvidia is trying harder, and Intel is experimenting, but do we have enough?
Naver (the South Korean search engine) is targeting other countries with a promise of localised LLMs. Also, Azure is giving prompting tips to biz owners using their platform.
Let’s get to it.
From our sponsor:
What if you had an AI Assistant to help you with your webinars & meetings?
Introducing Sessions Copilot, the world's most advanced AI webinar assistant: build your agenda, generate meeting titles and descriptions, AI-assisted polls, smart transcripts, and get the meeting summary in seconds.
Supercharge your webinars, workshops, and sessions with Sessions AI Copilot.
🤌 Our Picks
Highlights if you've only got 2 minutes
1/
While Microsoft and Google are playing the universal AI game, Naver is going down a different path. If you don’t know, Naver is the go-to search engine in South Korea, with more than 2/3rd of the users using it.
Their new ChatGPT-like systems are all about localised AI applications, focusing on countries with political sensitivities in the Arab world, and non-English speaking countries like Spain and Mexico.
This is a pretty savvy move if you ask me... Governments in these places are itching to have their own AI systems, ones that are tailor-made to their political and cultural contexts (and outside the US data control). For them, it could be a perfect fit, baby!
Unpaywalled version here.
2/
You need the pan to make a pancake, and in the AI kitchen, that we are running out of pans, aka the chips. Nvidia is the primary maker of these chips. They are trying to make more and better ones, but the demand is too high. This has left everyone, especially the smaller AI startups, scrambling for resources (or begging cloud giants for more rental space). WSJ article without paywall.
So, what's the fix? I wish I had the answer, folks, but this one's got even the top brass scratching their heads. Could that be Intel? They are working on VPUs, which they claim can sustain and offload the AI burden on CPUs, GPUs, and clouds. More on that here. One thing's for sure—whoever comes up with a solution is going to be sitting pretty!
3/
I’m a sucker for good dialogue in cutscenes in story games but they are rare. And it’s kinda boring to press E to interact with the character, then listen them ramble on with poor dialogue. But hold onto your joysticks, gamers, because Nvidia is cooking up a new suite of features for just that.
That's right, folks, we're talking about a real back-and-forth, heart-to-heart with your digital buddies. Imagine a future where you hold down a button, speak your truth, and get a response drafted by AI. These are custom responses based on your dialogue in the side characters’ voices that can tell you more stories, and unlock new missions, and much more. This will take story games to the next level of player-content interaction.
🛠️ Cool Tools
Product launches, updates and demos
Madlad- A Discord bot that can do anything a community moderator can. (link)
Chariot - The API for building LLM applications. (link)
Donut - OCR-free document understanding transformer. (link)
Stork - AI-based collaboration for hybrid teams. (link)
Dreamore AI – Interpret, paint and share your dreams. (link)
WP Wand – Create quality content 10X faster inside WordPress. (link)
Gptfile - A way to organize files with natural language using gpt-4. (link)
Prototyping games and experiences with ChatGPT. (link)
Podcast embedddings - 1000s hours of audio transcripts and serve 1M+ embeddings across the best podcasts. (link)
Clothing Segmentation on Replicate - This model can detect clothing using a custom state of the art clothing segmentation algorithm. (link)
Open source implementation of Character AI - Create any character into a discord channel and hangout with the character and your friends. (link)
CoachVox - Create an AI version of yourself to coach, mentor and answer questions just like you would. (link)
🤓 Miscellaneous
News, podcasts, videos, blogs etc
Photoswap - Personalized subject swapping in images. (link)
OlaGPT - Empowering LLMs with human-like problem-solving abilities. (link)
The fundamental quantities of LLMs - Compute. (link)
State-of-the-art fMRI-to-image approach that retrieves and reconstructs images from brain activity! (link)
Large-scale instruction tuning SFT data to equip LLMs with general tool-use capability. (link)
Production AI systems are hard. (link)
How to build AI products people want. (link)
A look at media coverage of generative AI over the past six months, as journalists and academics criticize the hype and provide ways to improve reporting. (link)
Nvidia's next DGX supercomputer is all about generative AI. (link)
AI will slightly drain the swamp of higher ed. (link)
AI server cost analysis – Memory is the biggest loser. (link)
Nvidia chief says with AI everyone now can be a programmer. (link)
🎓 Learn
How-to’s and resources
👋 Too many links?! I created a database for all links mentioned in these emails. Refer 1 friend using this link and I'll send over the link database.
📰 Unclassifieds
Short, sponsored links
HackAPrompt - Prompt hacking competition to enhance AI safety and outsmart large language models. Deadline: June 3rd.
Grow your business so fast it scares you using an “AI TV Channel” and Virtual Influencers at MoneyToaster.
Chroma and LlamaIndex are hosting a happy hour in SF on June 1st. Register here.
From book idea to Amazon listing in as few as 10 days — A.I. to write your entire book, then a roadmap to make it a bestseller.
Have a product, service, job, event, newsletter, app, book, movie, tool, or anything you'd like to share with over 100k subscribers?
Advertise
🖼 AI Images of the day
Funny memes and pics from around the web
Send this to 1 AI-curious friend and receive my AI project tracker database! Use this link.
How was today's email? |
Reply