- Born Analog//Living AI
- Posts
- AI Agents are here!
This past week has seen some amazing revelations about completely autonomous AI agents that may usher in a completely new way of doing work related tasks in the digital realm. Let’s dive in…
First of all though, meet Devin. He’s your everyday software programmer flogging his wares on the internet - in this case building websites of varied complexity:
All very un-interesting and normal, except that Devin is not a person but one of the completely autonomous agents I mentioned earlier. Following a simple instruction by Wharton Professor, Ethan Mollick (@emollick), to go on Reddit and offer to build websites, Devin overcame a number of challenges, reported back regularly on progress, asked qualifying questions before then finally deciding to charge for his services! Professor Mollick pulled the plug before anybody engaged Devin, but nonetheless, this is incredible (something I find myself saying more and more these days!)
Devin has been created by Cognition Labs. Here’s what they had to say about Devin’s creation and possibilities:
Today we're excited to introduce Devin, the first AI software engineer.
Devin is the new state-of-the-art on the SWE-Bench coding benchmark, has successfully passed practical engineering interviews from leading AI companies, and has even completed real jobs on Upwork.
Devin is… twitter.com/i/web/status/1…
— Cognition (@cognition_labs)
1:50 PM • Mar 12, 2024
Seems to me that we are opening up one huge can of worms with agents like this. What is next? If you assume that there will be a percentage of people who will look to deploy these increasingly powerful agents for nefarious means - there will be whether we like it or not - then we could be in for some weird rollercoaster of a ride in the coming years. This tool was built to augment to workforce (read: replace jobs) and that this will happen is somewhat inevitable.
A recent study has suggested that GPT-4 is significantly better at persuading humans than humans themselves when it knows some personal information about the person.
🚨LLMs outdebate humans when given access to opponents' personal Information!🚨
In a pre-reg study (N=820), participants who debated ChatGPT had 81.7% (p<0.01) higher odds of agreeing with their opponents after the debate (compared to a human baseline)
📜arxiv.org/abs/2403.14380
— Manoel (@manoelribeiro)
1:35 PM • Mar 22, 2024
Combine that power of persuasion with a completely autonomous AI agent and join the dots.
AI Agents are coming and we are nowhere near ready for them!
My working assumption on these agents becoming mainstream has just gone from two years to six months. These agents will come to dominate the digital ecosystem in the same way that humanoid robots like Figure 1 or Optimus will soon be ubiquitous in the physical realm.
“May you live in interesting times”. The irony of this blessing is that it can also be a curse. Scary times. Exciting times!
