The AI Bubble: Scaling Won't Save You

If you’ve been reading our newsletter, you know that we at Tune are AI optimists. However, we also keep our eyes peeled for reasoned opinions from the opposite side, like NYU Professor Gary Marcus’.

Is A(G)I worth all the hype?

Professor Marcus firmly says no and presents a series of fundamental problems that have gone unsolved for decades. He argues that AI still has major reliability issues when it comes to responses in unfamiliar areas of knowledge (i.e., stuff it’s not trained on).

Common-sense reasoning is still widely uncommon. And "we may have reached a period of diminishing returns of pure LLM scaling.” Where, then, can we extract real business value from?

Read more on Professor Marcus’ Substack

Making AI work for the enterprise

This skepticism toward the over-enthusiasm of AI startup founders is rooted in reality. “In no case could one simply use one of the foundation models straight out of the box and have it reliably solve business problems,” finds a recent article in Fortune.

So, what can you do to make AI work for you?

Tribal knowledge from the business leaders we work with shows us that enterprises can make the most of AI if they focus on:

• Use cases: Don’t let the employees figure out how to use the chatbot you’ve given them. Instead, prescribe and demonstrate use cases to improve adoption.

• Fine-tuning: Not everyone needs an LLM. Yep, you read that right. Use a small language model (SLM, if you will) and fine-tune it to produce better results.

• ROI: Exactly what improvement in efficiency, productivity or cost savings do you anticipate? Figure that out and build forward - incrementally.

There is a fantastic story about how Prosus, a European technology investment firm, uses its Toqan AI system across its portfolio companies. The results of sustained interest in optimizing AI systems are inspirational.

Read more here

Generative AI and the fabric of the internet

This week, Mat Honan, the editor-in-chief of the MIT Tech Review, claimed that AI “means the end of internet search as we’ve known it.” Essentially, he suggests that people will use services like ChatGPT or Claude for their search instead of Googling things as we do now.

In a way, it’s already happening. The man who blew up a Tesla Cybertruck outside the Trump Hotel apparently used ChatGPT for his planning. Lawmakers in the UK are cracking down on explicit deepfake images made with GenAI.

This evolution has had a profound impact on enterprises, content creators and users alike. Some of his observations:

• The nature of search queries will change. “You can describe what the bird in your yard looks like…and get an almost human explanation put together from sources previously siloed across the internet.”

• We’ll see a zero-click future. Meaning no one will search for something and land on your website. Traffic as a metric would be moot.

• Paywall is being breached: LLMs are using copyrighted material to recreate content and making it available for free!

If there is one long form you read this week, make it this terrific meditation on the future of Internet search on the MIT Tech Review.

Whether you’re an optimist, skeptic or realist, there’s one thing you’d agree on: AI is changing the world.

How are you changing with it? Let’s talk. Reply and we’ll set up a meeting to see what you can do with AI today!

Best,
Anshuman Pandey

P.S. Llama 3.3 70B is now available on Tune Chat