Introduction - the latest firestorm
A couple of days ago a new and exciting video dropped in the realm of AI. Cognition software released a presentation of their new product “Devin - the AI software engineer”
Cognition Labs join the club of many silicon valley AI companies that raised millions of dollars in the current AI goldrush. Without a doubt, investors are ready to bet on these guys.
Impressive
The demo shows how “Devin” solves a variety of problems that are usually done by human software engineers. For example it can autonomously find and fix bugs in codebases or complete real jobs on Upwork. Devin can even train and fine tune its own AI models.
The production value on the videos is also nice. The music and editing make you feel like you are being shown something amazing. Kudos for that.
The list of demos is varied and impressive. I mean this is what we do everyday as part of our jobs in the office isn’t it?
Is your job at risk?
One might think we enter the era when software engineers are no longer needed. On the other hand we can walk to the office by foot everyday instead of taking a bus or a car. That doesn’t mean that cars are taking over walking any time soon.
The tasks that are being presented are only a fraction of what we do at our job. Writing code and solving bugs can feel like this is all that we do at work - but it is not so.
I know that everyone hates having meetings but there is a reason we have them. There are dozens of individual decisions that a software engineer makes during the day. Starting from what task to tackle at a particular time to how to structure her code. AI cannot decide (at least not yet) which task is most pressing for the business without human intervention. And when it finishes, it cannot verify that whatever task is done is satisfactory and if there is no rework needed.
Should you be worried? There is the eternal question of progress. What is the acceptable rate? We heard for the last 10 years that self-driving cars are around the corner and they will eliminate drivers' jobs from the market. This has not happened yet, because self-driving is a much harder task to solve regardless of what Elon Musk will have you believe.
Final thoughts
The demo is really cool but we don’t know what is the API bill Cognition Labs are paying to run Devin for 40 minutes using however many API calls are needed to complete the tasks. We have no idea how expensive it is to run Devin. Maybe it is more expensive than a junior dev?
Demos are most certainly cherry picked to show the best results - maybe they ran 100 tries until getting an impressive demo
People love benchmarks, so the fact that the demo is viral will give Cognition Labs a nice exposure and more attention to the AI revolution.
However, if you go to cognition-labs’ website you can see that they are hiring a developer. Looks like we still need humans to write the next version of Devin.
ความคิดเห็น