AI/ML news summary: Week 47

Another week in AI means more breakthroughs, new models, incredible research, and massive leaps in hardware. I’ve scoured a lot of dark and obscure place of the interwebs to bring you this content as usual. And for the n@@bs that you are; I always try to make the nerdspeak a little less so y'all can pretend you are geeks !

Just sit back and relax a bit y’all because this week in AI was a lot quieter than usual. But it wasn’t without a few juicy bits though.

Remember last week’s talk about scaling compute hitting a wall (read: AI/ML news summary: Week 46 ). Well, the focus has shifted to inference-time compute scaling, and that is basically teaching AI to work smarter, not harder.

Meanwhile, some of the heavy hitters launched new models:

• Google’s Gemini: Took the crown in the Chatbot Arena like it was on a mission to outshine GPT-4o.
• Mistral’s Pixtral 124B: A new multimodal model that’s open-source and ready to rock.
• Alibaba’s Qwen2.5-Coder series: Ranging from 0.5B to 32B parameters, these models are practically screaming, “Hey devs, use me!”

But the real headline this week is the fact that you can now test yourself (or someone else) online through an ML supported test, to determine if you have a STD (or not, that can also be a possibility ). You just simply answer a few questions, and bingo! You have an STD!

Ummm… I think, that’s now how it’s supposed to work, right?

Anyway, if you want a scientific explanation for that itch you got, visit: ZIZ | Consult your browser will hopefully translate. ? Pro tip: open up the website in a private session, so as not to upset your partner more than you usually do ?

If you like this article and you want to support me:

1. Comment, or share the article; that will really help spread the word ??
2. Connect with me on Linkedin ??
3. Subscribe to TechTonic Shifts to get your daily dose of tech ??
4. TechTonic Shifts has a new blog, full of rubbish you will like !

What’s new in AI this week

1. Mistral’s Pixtral large and Mistral chat
2. Alibaba’s Qwen2.5-Coder series
3. ChatGPT outsmarts doctors
4. Nous Research introduces Forge reasoning API and Nous chat
5. Google’s Gemini surges past OpenAI
6. Emails expose OpenAI’s tumultuous early days
7. Amazon shows us some Trainium 2 chips
8. ChatGPT Desktop integrations launched
9. Largest Multilingual pretraining dataset released

Short Reads/Videos to pretend you’re working

1. What is Agentic RAG?
2. The future of programming: Copilots vs. Agents
3. Is AI progress hitting a wall?
4. OpenAI reveals new “Operator” AI Agent
5. A modern approach to causal inference

Repos & Toolsets

Not much this week…

1. Browser Use : Connects your AI agents with the browser. No biggie.
2. Perplexica : An open-source search engine alternative to Perplexity AI. You know, for when you want your searches your way.

Perplexica is kinda cool. For it to run, you need Ollama. This video shows you how to setup Perplexica on your own PC:

Top research of the week

Will all the lazy researchers raisy their hands?

1. LLaVA-o1 : Combines OpenAI’s reasoning with vision language models for better stage-by-stage interpretation.
2. Scaling Laws for Precision : Explains why training in lower precision can save resources without wrecking performance.
3. Garak: A framework for probing LLM security vulnerabilities. Think of it as AI’s red team.
4. Needle Threading : Evaluates LLMs on handling complex retrieval tasks. Spoiler: They struggle with long contexts.
5. AutoGen Studio : A no-code tool for building and debugging multi-agent workflows. Perfect for your next AI experiment.

Links you’ll like, ya nerd!

1. Nvidia Open-Sources BioNeMo Framework : Protein design, small molecule generation, and more, now open for biotech nerds everywhere.
2. Anthropic’s Prompt Improver : Automatically refines prompts for better results. The AI version of grammar check.
3. Elon Musk’s Antitrust Claims : Musk is suing OpenAI and Microsoft, accusing them of trying to monopolize generative AI. Popcorn, anyone?
4. Rabbit R1’s AI Interface Redesign : No, it ain’t dead, and yes, I got one. This device can completely redesign its interface now based on just prompts. Zelda-inspired UI, anyone?

And that’s the week in AI, decoded and snarkified for your reading pleasure.

Signing off - Marco

Well, that's a wrap for today. Tomorrow, I'll have a fresh episode of TechTonic Shifts for you. If you enjoy my writing and want to support my work, feel free to buy me a coffee ??

Think a friend would enjoy this too? Share the newsletter and let them join the conversation. Google appreciates your likes by making my articles available to more readers.