Reality Checking the Autonomous AI Bots

Reality Checking the Autonomous AI Bots

We love cheerleading the future at Meteor, but wow AI has a lot of cheerleaders these days, and they are feeding a hype cyclone.

Don't get us wrong, we absolutely believe in the power and long term impact of these tools, they are going to change the world. At the same time, social media hustlers are pouncing on the latest developments in a bid to bank followers and influence. It's hard to know what's real.

Which leads us to autonomous AI bots.

AI projects demonstrating new powers to autocomplete complex tasks have exploded over the past few weeks, thanks in large part to pioneering applications built on OpenAI's GPT-4 large language model (LLM), AutoGPT and BabyAGI.

These follow Microsoft's Jarvis and startup LangChain in creating environments where LLMs can feed outputs back to themselves to chain a series of prompts together, including prompts written by the models themselves, to iterate and improve at a task.

Put to use as an agent, people have managed to have AI order a pizza using Eleven Labs for voice generation and Twilio to manage the phone call; read about recent events and prepare a podcast outline; do research; and more.

It's easy to get the impression these AI agents, given the barest of instructions, can basically figure out everything required to get almost any result requested. For example, one highly shared Tweet described how AutoGPT was asked to build an app, and it went out and provisioned the required infrastructure on its own.

Already some people are talking about self-replicating super computer viruses, and the imminent achievement of the singularity, when an upgradable intelligent agent reaches runaway reaction mode. More prosaically, an entire genre of HustleGPT has spawned among influencers shilling ways to get AI to build entire businesses and do all the work while they kick back and rake in the profits.

We don't want to rain on anyone's parade, but while these explorations are clearly exciting, they are far from complete, and still very much experimental. When tech publication Ars Technica tested AutoGPT, it concluded the hype has raced ahead of the reality.

"They need a lot of human input and hand-holding along the way, so they're not yet as autonomous as promised. … In fact, the limited usefulness of tools like Auto-GPT at the moment may serve as the most potent evidence yet of the current limitations of large language models."

Others have found big gaps that limit their usefulness, at least for now.

None of this is to say the limitations won't be overcome. It's easy to imagine integrating a payments API to have an AI agent transact purchases on your behalf, for example, something that's not currently possible. But for now, take a chill pill and enjoy the ride.

Create

Untitled new work from Rikkar.

"Accumulation #2," part of the Mad World series by synthetic photographer CH D11.

"Being left behind" by Alizé Jireh.

Compress

Generative AI One Ups Napster
Fake Drake and The Weeknd Track goes viral. Instead of copying your favorite artists' music on illegal file sharing networks, you can now copycat their entire act. The music industry is not happy.

TruthGPT
If you're Elon Musk and you're announcing a new AI company devoted to removing bias, where better than Tucker Carlson's show?

Exchanging Words
The SEC has re-opened a proposal that would allow it to regulate crypto exchanges, the latest in a long series of actions aimed at reining in the industry in the U.S. SEC chief Gary Gensler testifies today at a high stakes Congressional hearing, and this is expected to be a hot topic.

Shaq Attack
Shaquille O'Neal has finally been served legal papers in a lawsuit alleging he and other celebrities improperly marketed crypto products. He had ducked process servers for three months before getting caught outside his home in Atlanta.

Trump NFT Sales Spike Around Arraignment Appearance
Although total wallet holders have fallen since the original sale of Trump cards, profits are up. nd it's more than enough to cover the hush money paid out Stormy. The former President has pocketed between $500K and $1M on licensing fees, reckons NFT data aggregator CryptoSlam, with recent gains pegged to his appearance in a Manhattan courthouse on criminal charges for improper use of campaign funds.

Software's Gutenberg Moment
Long, nuanced deep dive on the implications of AI for society and software coding. “It is an exaggeration, but only a modest one, to say that it is a kind of Gutenberg moment, one where previous barriers to creation—scholarly, creative, economic, etc—are going to fall away, as people are freed to do things only limited by their imagination, or, more practically, by the old costs of producing software.”

How did we do today? Love us or hate us, let us know.