Hacker Newsnew | past | comments | ask | show | jobs | submitlogin
Discuss HN: Will 100M token LLMs replace most programmers?
3 points by Mariciano on Aug 31, 2024 | hide | past | favorite | 6 comments
I recently had a conversation with a friend who's a Delphi developer on a large ERP project. He's been relying heavily on ChatGPT for SQL queries and specific functions, and now he's talking about fine-tuning it on their entire codebase. His claims about the future of our profession have left me unsettled, especially in light of recent developments. Magic, an AI company, just announced some significant breakthroughs:

Developed LLMs with 100M token context windows (~10M lines of code or 750 novels) Their LTM-2-mini model shows strong performance on a new, more rigorous evaluation method called HashHop Claim their algorithm is 1000x cheaper than Llama 3.1 405B's attention mechanism for 100M token context Partnered with Google Cloud to build AI supercomputers using NVIDIA's latest GPUs Raised $465M in funding, including $320M from new investors

My friend argues that with capabilities like these, AI could replace most programmers soon, leaving only a few to oversee the AI's work. He even suggested many intellectual jobs could be replaced very soon due to rapid AI progress. As someone studying to re-enter the field (Angular, JS, HTML, CSS), I find this both intriguing and concerning. What do you think? Is this a realistic assessment of where we're headed? How do you see the role of programmers evolving with these increasingly capable AI assistants? Could tools like this make programming more accessible, or will they reduce the need for human programmers?



True story from a non-programmer.

I used perplexity.ai to help me write a simple C# program to save all my Outlook emails as .msg files.

I was able to do it in about a day, which is an amazingly fast velocity, if you consider that I had nothing in terms of background.

Not suggesting this scales to non-trivial problems, but it actually created a working solution for me.


While AI tools like Ninja AI and OpenCraft AI can automate routine programming tasks, they cannot fully replace the nuanced problem-solving and system design skills that programmers provide. These AI assistants enhance productivity but still require human oversight for strategic decisions and creative solutions. As AI integrates more into development, programmers may shift towards roles that focus on directing AI operations and tackling complex challenges that require human insight. Thus, while AI will transform programming, it will not make human programmers obsolete but rather evolve their roles in the industry.


There may be a bit of a gulf between what they claim, and what they can realistically and financially provide.

I prefer to see long term real world results myself before I start to worry about any possible ramifications, myself.


> ...leaving only a few to oversee the AI's work.

What, exactly, would this entail, and how many is "only a few?"


If we had 100M tokens plus an LLM that is a few iterations beyond claude 3.5 opus, then sure.


No.




Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: