It wasn’t even a week into Devin’s¹ announcement when the hoopla around him (her/it/whatever…) ran amok (amuck). FUD, trolling, hype, the end of humans as software engineers, everything came out all guns blazing…as expected!
That small 1:50 video presented by Scott Wu caught everyone’s attention. After Devin took up the task (prompt) from a human to
“1. benchmark llama2 on 3 different providers: Replicate, Together and Perplexity. 2\. Figure out their API formats and 3\. write a script that sends the same prompt/params to all of them”
Devin went into autopilot mode. Devin’s output (a React application
— “using full styling as the visualization”
\- deployed on Netlify) is still there (not sure if it’s for posterity) — https://sparkling-faun-f82b48.netlify.app/
Naturally Matt Billman gave an honorary mention² to Devin
What if all of this was a run-up to an April Fool’s joke with that video being nothing more than an old Camstudio output of Scott Wu doing his homework assignment??!! 🤣 There is a likelihood/probability/chance of it too! Wouldn’t that be a slap-in-the-face for Hype mongers — almost reminiscent of the Sokal affair³? Wishful thinking, eh?
The fact remains that research into (and the application thereof ) AI/ML will continue. For now, it is better that positive sentiments like Human First: AI as a Force Multiplier⁴ continue to appeal to our bias.
A Devin as an AI based software engineer naturally translates to high cpu cycles for even basic tasks (Just look at how even a local, quantized llama2–13b would hit 600+% CPU utilization on your shiny 16 Core MacBook Pro M3 with GPUs enabled with your llama2.cpp experiments for even “the most basic” prompts). If Devin’s SaaS’d on the cloud, naturally that rings as “kaching” for the Cloud Service Providers and Hyperscalers.
Sustainability warriors — Where art’ thou?
AI Act Practitioners — Where would your conformity assessments put the likes of Devin (including the context in which Devin is put to use) at? \[“unacceptable”, “high”, “limited”, and “minimal or no risk”?\]
Automation proponents have that cheesy mantra of “Use the humans to focus higher in the value chain”. In that context, “Humans first” also might sound equally banal. Be that as it may, we all can stay cautiously optimistic. \[Do we really have a choice?\]
It would really be interesting to see how the Services sector promotes Devinisque approaches to large transformation problems and development of greenfield solutions among others. Basic GenAI getting passé with Devin becoming the hype du jour? What say you?
Developers and AI enthusiasts like Mckay Wrigley, Ethan Mollick, Andrew Kean Gao have also tweeted about their preview experiences with Devin.
In your own capacity, how would you use Devin or a team of Devins?
How would you monitor its/their work?
Explainability of blackbox like GenAI consumption is a pain as it is. Devin naturally cannot be a blackbox (was \[s\]he/it transparent enough in Scott Wu’s video?)
One thing’s for sure — Now is the best time to be a techie!! 😂
References:
\[1\] Introducing Devin, the first AI Software Engineer
\[2\] Matt Billman tweeting about Devin
\[3\] The Sokal hoax
\[4\] Beyond the Hype: Leadership and the Hard Work of Realizing AI’s Potential by Francisco D’Souza
\[5\] Mckay Wrigley uses Devin
#devin #devingossip #genaigrapewine #genai #hype #artificialintelligence #lostcommonsense #sokalhoax