Aye, I’d wager Claude would be closer to 58-60. And with the model probing Anthropic’s publishing, we could get to like ~63% on average in the next couple years? Those last few % will be difficult for an indeterminate amount of time, I imagine. But who knows. We’ve already blown by a ton of “limitations” that I thought I might not live long enough to see.
Participants only said other humans were human 67% of the time.
On the other hand, the human participant scored 67 percent, while GPT-3.5 scored 50 percent, and ELIZA, which was pre-programmed with responses and didn’t have an LLM to power it, was judged to be human just 22 percent of the time.
54% - 67% is the current gap, not 54 to 100.
Sounds like he needs someone with training to help him through retraining his behavioral/thought patterns, something a functional social system would provide if those were as common as comment culture.
I’ve consulted the oracles, and we’re going with 2006.
You’ve reminded me that I’ve been meaning to look more seriously at an Ultracortex.
It seems to me that we’ve reached a crossroads. I’ve been very aware of the data mining, garden walls, data trading, privacy violations, security issues, ownership issues, etc. - for roughly 30 years. I regularly make the choice to be exploited for the benefits I extract, largely because the data they’ve gotten from me thus far I don’t highly value. But the necessity to develop strategies to keep the devil’s bargain beneficial has reached a fevered pitch. I want to train my own AI and public AIs. I want to explore the vast higher dimensional semantic spaces of generative models without API charges. APIs are vanishing as we speak, anyway, companies fearful of their data being extracted without compensation. Can’t really sit on the Open/Closed fence anymore.
I understand this perspective, because the text, image, audio, and video generators all default to the most generic solution. I challenge you to explore past the surface with the simple goal of examining something you enjoy from new angles. All of the interesting work in generative AI is being done at the edges of the models’ semantic spaces. Avoid getting stuck in workflows. Try new ones regularly and compare their efficacies. I’m constantly finding use cases that I end up putting to practical use - sometimes immediately, sometimes six months later when the need arises.
Very interesting! I’ll check it out. Thanks for the recommendation.
I just meant I work for a corporation. I produce videos for marketing, been doing it for 25 years.
What tasks are you thinking about?
It saves me 10-20 hours of work every week as a corpo video producer, and I use that time to experiment with AI - which has allowed our small team to produce work that would be completely outside our resources otherwise. Without a single additional breakthrough, we’d be finding novel ways to be productive with the current form of generative AI for decades. I understand the desire to temper expectations, and I agree that companies and providers are not handling this well at all. But the tech is already solid. It’s just being misused more often than it’s being wielded well.
Don’t blame CEO tomfoolery on generative AI. Generative AI is amazing.
I am super impressed by Blender, but there’s not really a substitute for After Effects yet, open source or otherwise.
I literally read it to mean they’re starting to run out of women and children to kill.
Until they can distribute the training load of large models to consumer graphics cards (and do something like SETI@Home) it does seem like the benefit of distributed training isn’t enough to overcome the friction.
Y’all run out of Confederate flags AND No Step on Snek flags?
Like a decade ago?
The papers have a ton of practical info about feasibility, implementation, etc.
Eliza is not indistinguishable from a human at 22%.
Passing the Turing test stood largely out of reach for 70 years precisely because Humans are pretty good at spotting counterfeit humans.
This is a monumental achievement.