Weekly Link Roundup #14
Yes, I started numbering them. It makes them more collectable.
Bluesky opens up federation, letting anyone run their own server: This is important for social networking from a technical standpoint and I do think it’s interesting to see this space still trying to figure itself out on both technical and business model fronts.
Google Chrome’s new AI can finish your sentences for you: “Help me write” focuses on providing writing suggestions for shortform content, such as filling in digital surveys and reviews, enquiring about product information, or drafting descriptions for items being sold online.” » This will be fun. AIs will write surveys and AIs will fill them out.
These 3 companies show why the next AI wave won’t revolve around chatbots: A couple of quotes I really like “Chatbots are like a floor for AI’s usefulness, says Amelia Wattenberger, principal research engineer on the GitHub Next team…For most tasks chatbots are just not useful enough, she adds. “They’re generalists not specialists.” And this talking about just one of the tools in the article “Turn on Adept—it can function as a browser extension—and its multimodal AI model will monitor pixels on your computer as you navigate between various software programs. Crucially, it’ll learn from your behavior at work—the specific way you pay invoices, or how you create a sales lead, for example—and can then mimic those tasks automatically.” »If you’re in #learninganddevelopment what do you think this does to job guide creation or even onboarding new hires?
Former OpenAI Researcher Andrej Karpathy Unveils Tokenisation Tutorial, Decodes Google’s Gemma: I’ve said before that more people across the enterprise will have to become technically literate on how AI works, than ever before. So don’t roll your eyes at this tutorial. Here’s why - “We will see that a lot of weird behaviors and problems of LLMs actually trace back to tokenization. We’ll go through a number of these issues, discuss why tokenization is at fault, and why someone out there ideally finds a way to delete this stage entirely,” said Karpathy.” »How will you even know what questions to ask if you don’t have a minimal understanding on how this stuff works?
Scale AI to set the Pentagon’s path for testing and evaluating large language models: I hear you - you don’t work with DoD so why is this important? Here’s why - the U.S. DoD is an $800 million/year org with about 3.5 million people across the globe. It is the definition of an 800 pound gorilla. So what DoD does can have outsize impacts. So Scale AI was selected “to produce a trustworthy means for testing and evaluating large language models that can support — and potentially disrupt — military planning and decision-making…this new one-year contract will supply the CDAO with “a framework to deploy AI safely by measuring model performance, offering real-time feedback for warfighters, and creating specialized public sector evaluation sets to test AI models for military support applications, such as organizing the findings from after action reports.” Point: How Silicon Valley learned to love America, drones and glory.
Arc browser’s new AI-powered ‘pinch-to-summarize’ feature is clever, but often misses the mark: The tl;dr - good UI, bad UX
Nvidia Hardware Is Eating the World: It’s GPUs are driving AI and “Nvidia now accounts for more than 70 percent of sales in the AI chip market and is approaching a $2 trillion valuation. Its revenue for the last quarter of 2023 was $22 billion—up 265 percent from the year prior. And its stock price has risen 231 percent in the last year.” »Ask your AI vendor what their access to GPU capacity is. AND…in any gold rush, the people who really get rich are the ones selling picks and shovels: Charted: Nvidia shows how the AI boom is broadening.
Microsoft releases its internal generative AI red teaming tool to the public: This - redteaming - will is a requirement with enterprise-deployed AIs.
Meta Staff Found Instagram Tool Enabled Child Exploitation. The Company Pressed Ahead Anyway: This just is so bad. “A review by the Journal of some of the most popular parent-run modeling accounts on Instagram and Facebook revealed obvious failures of enforcement. One parent-run account banned last year for child exploitation had returned to the platforms, received official Meta verification and gained hundreds of thousands of followers. Other parent-run accounts previously banned on Instagram for exploitative behavior continued selling child-modeling content via Facebook.”
A former Gizmodo writer changed his name to ‘Slackbot’ and stayed undetected for months: How very Doctorow. “Hiding on Slack isn’t all that hard, apparently; you just have to pretend you’re a bot. That’s what IT Brew’s Tom McKay did when he left Gizmodo in 2022, and he went undetected by the site’s management for months.”
Opinion An ‘education legend’ has created an AI that will change your mind about AI: This » “For Khan, the new era is bittersweet. He has created a model for responsible AI — a chain that requires AI software makers to be humble and collaborative; subject experts to do rigorous testing and customization; and government entities to be responsible but open-minded. But in transitioning his life’s work to AI, he’s confronting his own obsolescence.” Nuance: We Tested an AI Tutor for Kids. It Struggled With Basic Math.
Groq lets you use multiple AI models quickly — here’s how: Want to test multiple LLM models? Here ya go.
How many news websites block AI crawlers?: This will influence the training set that LLMs have access to.
Stable Diffusion 3.0 debuts new diffusion transformation architecture to reinvent text-to-image gen AI: The article is fine on the Stable Diffusion front but it really shines when it cites resources for understanding the difference between some fundamental AI elements like transformers and diffusion models.
Techstars Seattle is shutting down as accelerator shifts focus to cities with more VC activity: This is sad for Seattle.
This old ‘Fast Company’ article is a touchstone for the AI deepfake age: Not bad - nailing this from 3 years out.
Slackers: A decade later, it’s clear that Slack changed work culture, even at companies that don’t use it: Lessons » “Most enterprise software is shitty. It’s clunky and unintuitive; the user interface is an afterthought. It seems obvious that the people who make these apps (hi Concur!) know that only a few people need to be convinced to make the purchase, and user interface matters a lot less than business-critical features. The average employee might hate the app, but that doesn’t matter. They’re going to use it anyway. That’s one reason why Slack, the workplace chat app that formally launched 10 years ago today, is so unusual. Slack was the rare piece of enterprise software that spread through word of mouth, because it was actually, you know, good.”
Another “patent troll” defeated by Cloudflare and its army of bounty seekers: …and there was much rejoicing!
Saw something similar on Twitter the other day, where someone recorded their screen and had Gemini create code for the process. Pretty eye opening.
Also interesting to see the comments in the Khan article. Wonder how wide spread that feeling will be as AI becomes more integrated into different systems.