Post AdO2Vf0lpnzKAAcppg by signaleleven@social.sdf.org
(DIR) More posts by signaleleven@social.sdf.org
(DIR) Post #AdNurgSgKA4bOkffto by simon@fedi.simonwillison.net
2024-01-01T00:08:18Z
0 likes, 0 repeats
To round off the year, I pulled together some notes on "Stuff we figured out about AI in 2023"We figured out a lot! https://simonwillison.net/2023/Dec/31/ai-in-2023/
(DIR) Post #AdNv4zdWLNwadH5Tpg by simon@fedi.simonwillison.net
2024-01-01T00:10:32Z
0 likes, 0 repeats
Here's the table of contents for my 2023 in AI round-up
(DIR) Post #AdNzNTH7Hp6daHfsjw by simon@fedi.simonwillison.net
2024-01-01T00:59:00Z
0 likes, 0 repeats
"The most surprising thing we’ve learned about LLMs this year is that they’re actually quite easy to build." https://simonwillison.net/2023/Dec/31/ai-in-2023/#easy-to-build
(DIR) Post #AdO2Vf0lpnzKAAcppg by signaleleven@social.sdf.org
2024-01-01T01:34:04Z
0 likes, 0 repeats
@simon thanks Simon, I've read a lot of your writing this year. I really enjoyed it!
(DIR) Post #AdO4QqHahUrDa4K6Ou by simonwiles@fosstodon.org
2024-01-01T01:55:42Z
0 likes, 0 repeats
@simon Surely the issue isn't the engineering or the cost of the GPUs -- the "[i]f you can gather the right data" part is the part that means that control of these technologies remains (and will remain) in the hands of the huge corporations, isn't it?
(DIR) Post #AdO4rc0dsPdv8DA0Rs by simon@fedi.simonwillison.net
2024-01-01T02:00:21Z
0 likes, 0 repeats
@simonwiles There are some very extensive openly available training datasets these days - RedPajama is one example, and there have been more like that created since https://simonwillison.net/2023/Apr/17/redpajama-data/
(DIR) Post #AdO54kpqNEGQGjNOGe by simon@fedi.simonwillison.net
2024-01-01T02:01:01Z
0 likes, 0 repeats
@simonwiles This is one of the big questions I have around the New York Times lawsuit - if it succeeds and sets a precedent we may see this kind of training data become much harder to obtain for groups that don't have serious money to spend on licensing fees
(DIR) Post #AdO6Wa6NyNUx2iwjiq by dcreemer@sfba.social
2024-01-01T02:19:03Z
0 likes, 0 repeats
@simon Thanks for all of your work and writing on this topic. It's helped me tremendously!
(DIR) Post #AdOWj20OeUONrok5Gy by peter@thepit.social
2024-01-01T07:12:43Z
0 likes, 0 repeats
@simon this is a great roundup!
(DIR) Post #AdOd6SH3kchPiv5Yv2 by simon_brooke@mastodon.scot
2024-01-01T08:24:11Z
0 likes, 0 repeats
@simon the difference being, of course, that a suspension bridge – if built in a suitable place – is useful.
(DIR) Post #AdPHBC1ja6dCwqjb5U by simon@fedi.simonwillison.net
2024-01-01T15:53:10Z
0 likes, 0 repeats
@simon_brooke I've been finding useful applications for ChatGPT on an almost daily basis since it came outTwo easy examples: It's the world's best "what word do I need here / can I use instead?" thesaurus, and if you have an error message from anything it can get you started figuring out the problem faster than anything else
(DIR) Post #AdPJcMujPZIoROXqyG by simon@fedi.simonwillison.net
2024-01-01T16:20:28Z
0 likes, 0 repeats
A few people have pointed out that I didn't cover AI topics outside of LLMs in my postI originally planned to, but I also really wanted to publish it in 2023 and I ran out of time!Might need to do a "part two"
(DIR) Post #AdPMx9bCPYgz5Z0n1k by simon@fedi.simonwillison.net
2024-01-01T16:57:53Z
0 likes, 0 repeats
@bob_zim I think it's pretty surprising that you can get a system with this many capabilities based on so little actual code - if you showed me GPT-4 a few years ago I'd assume it was millions of lines of code written by thousands of engineers for over a decade
(DIR) Post #AdPP2Gc1xkmwCFxzCC by lewiscowles1986@phpc.social
2024-01-01T17:21:04Z
0 likes, 0 repeats
@simon You cover so much, if folks want non LLM AI, maybe they should write it themselves.
(DIR) Post #AdPeFhHbc0y1zVH6iu by simon@fedi.simonwillison.net
2024-01-01T20:11:35Z
0 likes, 0 repeats
@bob_zim might not be surprising to you, but it was definitely surprising to me!