weekend ai reads for 2024-06-28

June 28, 2024

📰 ABOVE THE FOLD: IMPACT ON CREATIVE FIELDS

When problems arose – from broad ones, like characters randomly switching sex, to more nuanced issues such as opaque motivation – he asked the programme to go back in and rework what it had come up with.

“Some creative jobs maybe will go away, but maybe they shouldn’t have been there in the first place,” the CTO said of AI’s role in the workplace. “I really believe that using it as a tool for education, [and] creativity, will expand our intelligence.”

  • NBC begins to ruin the Olympics earlier and earlier, don’t they?

 

📻 QUOTE OF THE WEEK

You think it’s cool to hate things. And it’s not. It’s boring.

Talk about what you love and keep quiet about what you don’t.

Liberal Arts (2012) (source)

 

🏗️ FOUNDATIONS & CULTURE

On Claude 3.5 Sonnet / Zvi Mowshowitz, Substack (sorry) (19 minute read)

The review by UK’s AISI is very good news, especially after Jack Clark’s statements that making that happen was difficult. Now that both DeepMind and Anthropic have followed through, hopefully that will put pressure on OpenAI and others to do it.

By creating a space where AI-generated content can be easily edited, refined, and incorporated into ongoing projects, Anthropic is bridging the gap between AI as a tool and AI as a teammate. This shift has the potential to revolutionize knowledge work across industries.

AI Survey: Four Themes Emerging / Bain & Company (7 minute read)

Concerns about security and conversations around implementation are more deliberate and informed as companies have a better understanding of the challenges based on learnings from their pilot programs. Concerns about organizational readiness grew while those around quality and risk have declined.

The CPG sector faces some unique challenges. The proliferation of data, for example, and its complexity—sources are scattered across retailers, suppliers, manufacturers, and consumers—have created massive issues in terms of harnessing the data to find, track, and capture value. At its core, the reason for this low success rate is that companies fail to perform the deep organizational surgery required to affect the broad-based change that’s needed. It’s never “just tech” when it comes to successful digital and AI transformations. Companies need to rewire how they work.

I Will Piledrive You If You Mention AI Again / Ludic Maratoa, Ludicity (23 minute read; star of the week so totally worth it)

Unless you are one of a tiny handful of businesses who know exactly what they're going to use AI for, you do not need AI for anything - or rather, you do not need to do anything to reap the benefits. Artificial intelligence, as it exists and is useful now, is probably already baked into your businesses software supply chain.

Generative AI is most helpful when assisting humans by solving problems where solutions are Hard to Make, but Easy to Check (HMEC).

One thing is becoming quite clear in this space of “Let’s make an LLM generate DATA for us” efforts – regardless of whether it actually works on a conceptual or practical level, people who make decisions will most definitely make attempts to apply the technology to every domain they can. Stopping them ahead of time is likely an exercise in futility until they see it break themselves.

 

🎓 EDUCATION

Conversation with Edison Durán Lucena: Enhancing Education with AI-Powered Books — EdTech startup secures $16,000 prize, plans to revolutionize publishing with augmented reality and AI for K-12 education. / Contxto (5 minute read)

Akadimia — Step into the future of learning with Akadimia AI, an immersive educational platform that brings history’s icons as your AI mentor

  • iOS only

  • character.ai with historical people like MLK & Dalí to support interactive learning (?)

Coach — AI-powered career development for every learner

  • collegiate (?) and career pathways and seemingly designed for pre-postsecondary students

AI Tutors Don't Know When to Stop Shutting Up — The chatbots don't know when to start chatting with learners. / Dan Meyer, Substack (sorry) (5 minute read)

  • mostly describing a UX problem

 

📊 DATA & TECHNOLOGY

Frontiers in synthetic data / Nathan Lambert, Interconnects, Substack (sorry) (13 minute read)

Correlations are also beginning to emerge between organization size, revenue, and rates of AI deployment—though the largest and highest-grossing organizations aren’t necessarily the most successful early AI adopters. Rather, the right blend of raw investment power and an agile data strategy seems to be the single greatest predictor of success.

In other words, if we rely on black box systems, thinking the error percentage looks pretty good while having no idea what that error percentage covers or how the system even works, then we owe it to ourselves to understand just how big the risk potentially is.

Announcing the AI Forecasting Benchmark Series / Metaculus (8 minute read)

Drawing on our experience with live agents in production, we distilled the requirements for a realistic agent benchmark to three key points. First and foremost, most real-world settings require agents to interact seamlessly with both humans and programmatic APIs over long horizons, in order to incrementally gather information and solve complex problems.

Furthermore, beyond their benefits in quality, when you build and deploy the RAG fine-tuned models in the Together platform, they also offer significant economic advantages, up to 150x cheaper, while it’s 3.7x faster during inference.

AI-Flow — Open-source platform for creating custom AI tools through a simple drag and drop interface, designed for innovators and creators

  • quite useful; complicated to setup locally

AI scaling myths — Scaling will run out. The question is when. / AI Snake Oil, Substack (sorry) (12 minute read)

 

🎉 FUN and/or PRACTICAL THINGS

How to use AI prompt chains for better results / Geeky Gadgets (9 minute read)

At their core, prompt chains are carefully crafted sequences of prompts designed to break down intricate processes into manageable steps. By decomposing a complex task into a series of interconnected prompts, each building upon the output of the previous one, prompt chains enable the AI to generate more coherent and contextually relevant outputs.

  • useful prompting advice

Pixlyze — Detect AI Images

Our CNN-based models demonstrate an accuracy rate of up to 68.9%, outperforming or matching human observers in many instances. Intriguingly, both the machine learning models and human observers exhibited a shared propensity to more effectively identify negative affective states, which may be attributed to the more intense and straightforward expression of these states.

  • having played tennis semi-competitively, “expressive behavior” is rarely vague, particularly for negative affective states; the accuracy rate seems really low for both models and humans (source: observed many racquets being broken mid-set)

  • regardless, the techniques used are interesting

Ario — a consumer AI assistant that initially focuses on the everyday tasks that are universal to so many of us: managing time, schedules, and family obligations

  • iOS only

  • free; and no data collection

 

🧿 AI-ADJACENT

Asking Al to generate the Tour de France — Nailed it. / crit.meme, Instagram (sorry) (1 minute video)

  • reminder: the Tour de France starts tomorrow, 29 June, in Florence

 

⋄