I wrote a thing about #AI for a local newsletter with a few thousand subscribers. So I kept it as non-technical as possible for that audience. I should say, after 35 years in the technology sector, I am a tech journalist/writer - not an AI scientist or programmer. Companies have hired me in the past to ‘explain’ complex tech, so let me know if I got this right. There’s a second, more technical post coming that explores the evolution of AI through neural networks to machine learning & intelligence. Hey, it’s the topic of the day, how could I not write about it? That’s kinda what I do. #artificialintelligence#essay 1/2
@CharlieMcHenry Anyways I think your piece wildly over-anthropomorphizes current AI systems. They don't "reflect" or "research", ChatGPT is a linear algebra function with mechanistic outputs. A statistical model of human language. Both have neurons and show some intelligent behavior, but a human isn't a transformer model.
@CharlieMcHenry
People may be talking about it. but that doesn't make it real.
The thing being pushed as AI is "only" the mechanisation of creating a plausible answer. Automated Populism - as if we need any of that.
Combined with photorealistic hallucinations, it is very likely to destroy the usefulness of the internet; the rapid evolution of fake news right before our eyes. Very singular, indeed ☹️
@CharlieMcHenry "Most experts just a year or so ago expected it to take ten years plus to get to AGI, but recent developments put us almost there"
This is dubious at best. AGI has been a "10-20 years away" technology since the Perceptron. Honestly I find it impossible to separate AI reality from AI hype, since both AI and Intelligence are still poorly defined and understood even less.
@Woodhouse3D We have a benchmark for that… the ‘massive multitasking language understanding’ (MMLU) measure/scale. Competent humans generally score @ 89% on that, and properly prompted, GPT-4 has gotten pretty darn close, in the low 80’s recently.
@CharlieMcHenry Oh, so they invented a test that "measures intelligence". Since ChatGPT scores so well it must be intelligent too. I see no problems with this.
2/2 in my second post/essay on #AI I’ve taken a quick look at the history and evolution of #artificialintelligence, dropped some names, and dived a little deeper in artificial neural networks (not as deep as transformers yet, that will come later). Again, the caveat hers is that I’m a tech writer/communicator - not a data scientist, programmer or network manager, so cut me some slack. This essay was written for my curious, non-techie but smart friends who seem to appreciate my efforts to break it down for them. There are far more detailed and technical pieces out there, but most would be beyond the average reader. https://docs.google.com/document/d/1NUZxWypcNJSkXLVP8MEdw0oFfuPOiFDDYatiMGeo2lA/edit
Add comment