You are currently viewing SemiWiki as a guest which gives you limited access to the site. To view blog comments and experience other SemiWiki features you must be a registered member. Registration is fast, simple, and absolutely free so please, join our community today!
While many professions will adopt AI/ML and the massive efficiencies it will bring as it learns more and more as it is used, many professions will consider it a threat. What do readers think the early adopter professions will be and those that will fight it tooth and nail?
Feel confident your research is accurate and complete while finishing your work faster than ever before.
legal.thomsonreuters.com
Software engineering. IMO, you really can't be competitive as a software engineer anymore unless you're using AI as an assistant.
Healthcare, especially radiology. Revenue generating tools already available. It still takes a radiologist to review the conclusions of the AI tools, but the tools increase productivity.
Drug research and development. In progress. "Computational pharmacology and toxicology." Many self-described experts think this is where the real money is. Too far away from my expertise to agree or disagree.
OpenAI says that ChatGPT is processing over one billion queries per day. The calculations in the gas tank link make no sense in that context.
I don't know how to equate the tests described in the arcprize link to common programming questions. I think they're probably not relevant to this thread. In a few test programming questions I submitted to ChatGPT the answers returned in a couple of seconds, so I doubt the cost was high at all.
Using AI for starting points in programming is very controversial in the software development community. Old-timers contend it's making people entering the field dumber. I think the opposite is true - using GenAI accelerates learning. You can't please everyone. Most modern programmers would be lost using assembler language. Technology moves on.
Any activity where actual problem solving (rather thean just talking about it) and cost and time to market are paramount. Which includes activities like ours (electronics, software) which aren't highly regulated and often aren't considered to be professions. The more cut-throat/competitive parts of finance/banking (again, these are - rightly or wrongly - less regulated in practice).
OpenAI says that ChatGPT is processing over one billion queries per day. The calculations in the gas tank link make no sense in that context.
I don't know how to equate the tests described in the arcprize link to common programming questions. I think they're probably not relevant to this thread. In a few test programming questions I submitted to ChatGPT the answers returned in a couple of seconds, so I doubt the cost was high at all.
Using AI for starting points in programming is very controversial in the software development community. Old-timers contend it's making people entering the field dumber. I think the opposite is true - using GenAI accelerates learning. You can't please everyone. Most modern programmers would be lost using assembler language. Technology moves on.
It is specifically about o3 model. Which requires orders of magnitude more tokens to process single query. Regular chatbots often rolls back to older models or even simple searches. I assume this is kind of difficult to benchmark and get similar evaluations from independent sources... Then there is working in context, such as existing codebase. Which also eats lot of tokens...
I tested GPT4/Claude/o1 and it works great in isolated scripts. I got mixed result in C/C++ (It solved semester worth of assignments in few minutes but struggled when i made even minor changes in requests). And it really struggled in Verilog.
It is specifically about o3 model. Which requires orders of magnitude more tokens to process single query. Regular chatbots often rolls back to older models or even simple searches. I assume this is kind of difficult to benchmark and get similar evaluations from independent sources... Then there is working in context, such as existing codebase. Which also eats lot of tokens...
I tested GPT4/Claude/o1 and it works great in isolated scripts. I got mixed result in C/C++ (It solved semester worth of assignments in few minutes but struggled when i made even minor changes in requests). And it really struggled in Verilog.
Agree about C/C++ code. You really need to be an expert in the field to judge the efficacy of any query result from GenAI, IMO. But it can still give a good thought-provoking answer in many cases, even in the early stage of AI development we’re at.
As for lousy results with Verilog, I’m not surprised a bit. The amount of public domain Verilog must be tiny for training purposes compared to the amount of C/C++ (or Python, or any other software language).
While many professions will adopt AI/ML and the massive efficiencies it will bring as it learns more and more as it is used, many professions will consider it a threat. What do readers think the early adopter professions will be and those that will fight it tooth and nail?
I think it will be easier to list the professions that will not adopt AI but I cannot think of one off hand. Everyone will claim to be adopting AI to stay competitive. AI will touch every profession, one way or another, you may know it or not. Hopefully AI can cut out the middlemen in healthcare, legal, tax, and insurance industries. The bureaucracy of it all is overwhelming.
arcprize.org
analyticsindiamag.com
