LLM links

Ethan Mollick on Devin; The Zvi on Devin; Suleyman's Inflection Exit; Mollick on treating LLMs as human;

Ethan Mollick writes,

If you are used to chatbots, working with Devin can feel like seeing the future. The interface is radically different, more like managing a project than prompting an AI. When given a task like “create a web page that lets me see the distance between airports.” The first thing Devin does is make a plan, listing the research it will do and the coding steps it will take. It then executes it autonomously, searching the web for sources of airline data, downloading it, and building the program, including debugging the results.

…It feels like using a contractor, rather than a chatbot.

Zvi Mowshowitz writes,

It is clear that Devin is a quantum leap over known past efforts in terms of its ability to execute complex multi-step tasks, to adapt on the fly, and to fix its mistakes or be adjusted and keep going.

The Zvi emphasizes the dangers of recursive self-improvement, of asking Devin to build a better Devin.

TechCrunch reports,

Microsoft has hired Mustafa Suleyman and Karén Simonyan, co-founders of high-profile AI startup Inflection AI, and several of their colleagues in one of the strangest deals

Strange indeed. Ordinarily, you would expect Microsoft to acquire Suleyman by buying Inflection. For the founders to jump ship, leaving their investors with nothing, would be immoral, incur reputation risk, and probably be a breach of fiduciary duty (although just about every founder I’ve ever invested in acted as if there was no such thing as fiduciary duty).

But then the story says that Reid Hoffman is a backer, and he says it’s all good.

Hoffman said all investors in Inflection AI “will have a good outcome today, and I anticipate good future upside.”

And Tyler links to a tweet by Sheel Mohnot that explains it this way:

Microsoft is making the Inflection shareholders whole via a licensing deal… and they still keep their shares in Inflection, which is an ongoing concern. This whole thing is a regulation dodge to get around the sad state of M&A at the moment.

So it’s a merger, but to avoid the Wrath of Khan it’s not a merger-merger.

Ethan Mollick writes,

the best users of AI are often managers and teachers, people who can understand the perspective of others and correct it when it is going wrong.

…Telling the system “who” it is helps shape the outputs of the system. Telling it to act as a teacher of MBA students will result in a different output than if you ask it to act as a circus clown. This isn’t magical—you can’t say Act as Bill Gates and get better business advice or write like Hemingway and get amazing prose —but it can help make the tone and direction appropriate for your purpose.

…treating AIs like people seems like an inevitability, so figuring out how to do it in safe, productive ways may be better than the alternatives.

It strikes me that the best practices involve treating the LLM as a talented but inexperienced individual, useful but in need of guidance. The worst practices involve treating it as omniscient.

substacks referenced above:

@

On the necessity of a sin

@

On Devin

@

Which AI should I use? Superpowers and the State of Play