Just posted a blog post on the AI hype like late 1...
# 05-ai-newsc
Charlie Graham
09/18/2023, 9:09 PMJust posted a blog post on the AI hype like late 1990's feedback apprecaited! https://twitter.com/imcharliegraham/status/1703876476423356922
a
Andreas Ramos
09/19/2023, 2:15 AMin 1994-6, people didn't "quickly realize" anything. it took a long time. it was 98-99 that it began to really grow.
➕ 1
Andreas Ramos
09/19/2023, 2:18 AMthe dotcom boom crashed because there was massive fraud. traffic was wildly exaggerated, both future traffic and actual traffic. investors realized most of it was fake and they pulled out. however, good companies also crashed.
Andreas Ramos
09/19/2023, 2:20 AM(I worked through all of this. came to SV in 1992. bought my house in Palo Alto in 98. worked at many dotcoms.)
Andreas Ramos
09/19/2023, 2:25 AMtoday...similarities. dotcom boom all over again. energy, excitement, an entirely new kind of thing, huge potential.
However... this time it's much faster. HTML appeared march 93. it was in 95-96 the first serious investing and the public began to notice. three years. ChatGPT exploded into the world, like Minerva. we've gone through five years of dotcom evolution in three months. VCs are not going to sleep this time. nearly every VC fund has money in AI.
🎯 1
Andreas Ramos
09/19/2023, 2:30 AMHowever, AI so far is LLMs. there are very few consumer services. the vast majority of users don't understand how to prompt. any mom could use Yahoo! in 1996. OpenAI doesn't bother trying to reach the public.
either a useful consumer product appears in 2024 or the public will lose interest.
c
Charlie Graham
09/19/2023, 5:28 AMThanks for your take and sharing your perspective @Andreas Ramos! I was a 96 undergrad at Harvard and grew up in the SF Peninsula. I remember the exact moment of downloading the first image on Netscape in '94/'95. It was a game changer for me and all of my roommates and CS buddies. There was palpable excitement.
And then I co-founded a chat co-browsing company 1998 in the bay area that lasted until 2004 so I also had a front seat on all of this.
IMHO there was definitely fraud (Worldcom, etc.) but the dotcom crashed as everyone was going after "eyeballs" instead of real revenue and few consumers were willing to spend money on the web as there was too much fear of inputting in your CC. At some point investors realized there was no profitable model and the market crashed. Fraud also contributed but I didn't remember Fraud crushing pets.com, Webvan or most of the big names during that era. I felt was more imho that they ignored profitability as a metric and that the infrastructure/userbase was not there to make the market and cost structure match their aspirations. I could be mis-remembering though.
Charlie Graham
09/19/2023, 5:39 AMAs for LLMs - I may be wrong but I believe LLMs have been around for a few years? ( Using LLMs to write blog posts/marketing content has been around since 2021 and before then if you include GPT2). Companies like copysmith.ai, jasper, type.ai were all built on that. (I thought about building one of those in April 2021 :)). GPT4 is definitely a huge improvement though.
However, playing with GPT 4 a lot I believe GPT has big limitations that people are not realizing unless they use it very frequently.Tell it to write 100 children's stories and use whatever prompts you want and you'll start seeing it becoming repetitive. (I stopped after about 30). It feels like it gets to a local maxima of what it thinks is the "best children's story" . (And you can find similar limitations with code, answering questions etc).
💯 1
Charlie Graham
09/19/2023, 5:43 AMChatting with a lot of people starting AI companies based on LLms, I found it super interesting that very few are in production mostly because of accuracy issues (They can't get it accurate enough to sell yet). They can demo it but can't seem to nail it 98+% of the time which is often required for it to work at scale. The ones that are live are the ones where customers are ok with errors, or where it is just used as a "suggestion".
It feels like we are still far off from getting a high enough accuracy for a lot of use cases but I could totally be wrong.
a
Andreas Ramos
09/19/2023, 5:31 PMthe fraud was the wildly exaggerated future traffic. hockey sticks! investors believed that.
it wasn't basic fraud like Worldcom (there was lots of that), it was carnival barker fraud of impossible promises. remember Webvan?
I worked in startups where NYC and LA bizdev people came in and announced traffic projections and the engineering team would burst out laughing.
it's happening again. Sam, Peter, and others declare AGI will replace 50% of all jobs, etc. AGI doesn't even exist yet.
c
Charlie Graham
09/19/2023, 5:35 PMAh. So totally bogus projections? I don't consider that fraud as much as totally irrational projections based on everyone thinking the technology would progress faster than it did. I mean Webvan was right (Think Instacart/Amazon Fresh), Pets.com was right too. Their projections were just 5-10 years early.
Charlie Graham
09/19/2023, 5:36 PMI mean I would argue we have the same thing with AI companies - not just AGI but almost all AI companies rn thinking AI will progress at the same rate it did between GPT 3 and 4. For example that eveyrone will use an AI Agent for travel instead of Google Trips /Expedia in the next 2 years.
Charlie Graham
09/19/2023, 5:37 PMOr that GPT will replace Google search in the next year.
Charlie Graham
09/19/2023, 5:37 PMIt proves my article! 🙂
a
Andreas Ramos
09/19/2023, 5:56 PMyes, very good example: Microsoft believed LLM would replace Google search. even Google believed it. now? v unlikely.
Andreas Ramos
09/19/2023, 6:02 PMthere was no web Analytics back then.
we counted server hits.
A web page with one cat pix was two hits.
it took only a short while for some to figure out that a photo could be sliced into ten slices. Bingo! ten hits! OMG! eyeballs! traffic spikes!
Why not one hundred slices? and 20 pixs on a page?
Wall Street went bonkers.
all because they didn't understand hits. I still hear this today.
👆 1
c
Charlie Graham
09/19/2023, 10:41 PMHah. I did not know that. Thx for sharing. I mean it's not much different than today's "ad hits" where the ad is below the fold I guess. People gaming the system. Did Webvan or Pets.com do that or just other more shady sites?
j
Jocelyn
09/21/2023, 6:22 AMinteresting conversation. all else aside, I do agree that the zeitgesit does feel eerily reminiscent of the first boom, although as this conversation demonstrates its sort of a subjective as to how far that analogy reaches once you get into the nitty gritty. The feeling is definitely there though
c
Charlie Graham
09/25/2023, 10:18 PMor as supposedly said by Mark Twain, "History never repeats itself, but it does often rhyme"