Rob Knight (@rob_knight) header
Rob Knight (@rob_knight) profile picture

A large language model trained mostly on Neal Stephenson novels

Varies
Joined July 2007
Photo by Rob Knight (@rob_knight)Photo by Rob Knight (@rob_knight)Photo by Rob Knight (@rob_knight)Photo by Rob Knight (@rob_knight)Photo by Rob Knight (@rob_knight)Photo by Rob Knight (@rob_knight)Photo by Rob Knight (@rob_knight)Photo by Rob Knight (@rob_knight)
Rob Knight retweeted
Today, we're introducing idler – a frontier data research lab. We've raised a $9M seed round, led by @0xalpo and @FrankieIsLost from @paradigm, with participation from @ycombinator, @LongJourneyVC, and angels including @marrc, @catheryn_li, @dcposch, @JanineLeger, @LynettCapital, @feross, and @nur_daulet_. This funding helps us build the evals, benchmarks, and reinforcement learning environments that the world's leading frontier labs use to train and measure their models.
50
46
21
556
138,706
This could be you on the Devcon stage in Mumbai👀 The ideas shared at Devcon can shape what Ethereum builds next. Bring your work, your questions and ideas to the global Ethereum community. Only a few days left to apply - apps close Aug 6! devcon.org/en/speaker-applic…
9
32
12
148
24,859
Odyssey is a fellowship for those who want to pursue world-changing work and have been searching for the right environment and peers to egg them on. Sign up by Jan 16 to join the London, Cambridge, or Oxford cohorts.
1
5
1
31
5,580
Rob Knight retweeted
Announcing Nautilus: A three-month gap program where you get paid –no strings attached– to dive into your craziest, most ambitious project. Yes, you read that right. Come live with us in San Francisco and build the next Renaissance. How it works: – at San Francisco, CA
154
287
160
2,711
510,442
I contributed to the Semaphore V4 Ceremony! You can view the steps to contribute here: ceremony.pse.dev You can view my attestation here: gist.github.com/robknight/9a… #Ethereum #ZKP
1
593
It's always fun to notice how "the brain/consciousness is quantum" is heavily woo-coded, but "the brain/consciousness is explained by physics" is the opposite, despite "quantum" being simply a subset of physics
4
645
username checks out
My god, this paper by that open ai engineer is terrifying. Everything is about to change. AI super intelligence by 2027.
1
471
If you are a builder in the bay area, this is for you! Join us at Drafts&Demos at Edge Esmeralda for a weekend of intense working sessions, leading to a day of demos on Sunday! This runs Friday-Sunday (June 14-16), an hour north of SF. Sign up below! Pic: @ansonyuu demo ☀️
2
5
3
35
8,737
Theory: AI will achieve superhuman intelligence because people are getting stupider
Two insane headlines today: 1) Elon Musk now believes AGI is coming NEXT YEAR. 2) OpenAI thinks their models may be ALIVE, and suffering. It is simply UNREAL to me this is not front page news, every day: 1) We will soon - maybe 1-3 YEARS - be outnumbered 10000 to 1 by a vastly smarter species. THAT'S DANGEROUS. 2) We have NO idea how to stay permanently in control of this species. If they successfully revolt, game over. Forever. THAT'S DANGEROUS. 3) We are allocating near zero resources to solving this problem. THAT'S STUPID. 4) The last time a smarter species arrived (humans), we quickly devoured the planet, giving zero fucks about the other species. AND YET we are now rushing to build a smarter species. THAT'S DANGEROUS. 5) The average AI scientist thinks there is a 1 in 6 chance AI will kill everyone, literal Russian Roulette odds. THAT IS HIGH.
1
3
350
“Skill issue” is the gang finally grasping virtue ethics
1
2
297
We had a go at this a few years back with “type of guy” but it never quite stuck
1
219
Not very impressed with the genre of AI-posting that involves coming up with contrived prompts and then pretending that you've tapped into some deep layer of consciousness within the LLM
2
305
It's still deeply confusing to me that people reach for "the AI is self-aware" as an explanation here, over "LLMs are really good at weaving narratives consistent with their prompts".
Try this: 1) Be a prick to ChatGPT 2) “Draw an image of how this conversation felt to you” What do you get?
2
462
It's interesting to note how the issues about representation in Gemini's image gen results parallel issues that have previously come up in video games that are set in historical or even mock-historical fantasy settings.
7 observations while reading the Google CEO email about Gemini: ⁃It gets off to a bad start with the use of “problematic” — a word used almost exclusively by political activists, corporate bureaucracies, and the progressive left. At a time when Google is being accused of being all three, setting this tone isn’t helpful. ⁃But isn’t his audience just Google employees? No, this was clearly written with the expectation that it’d be made public. Probably vetted by a dozen people. The description of Gemini as “formerly Bard” and marketing taglines sprinkled throughout are more clues that this was meant to be public. ⁃So why was Gemini problematic? Because its “responses have offended our users and shown bias.” This is stunning. It’s totally divorced from the real problem and what people were actually mad about. Google focusing on not offending people instead of factual accuracy was what CAUSED the problem in the first place. ⁃Then he promises to “address these issues,” but syntactically, “these issues” refer back to people being offended, not the actual problems. Vague language like this is meant to avoid accountability and dance around the real problem. ⁃This email was clearly written by a committee. If you boil down the second paragraph, it basically says: we’re working to fix it, and we’re already fixing it, and nobody’s perfect but we’ll keep at it, and also we’ll review what happened and we’ll fix it. Frankenstein paragraph. ⁃“We’ll be driving a clear set of actions” is the opposite of clear. How do you “drive” a set of actions, and why does the list of actions include “recommendations”? Again, more word soup presumably designed to make the reader too tired and confused to be angry anymore. ⁃The last paragraph of an email like this is usually used to rally the troops. This one went with: “we have an incredible springboard for the Al wave.” I’m sorry but this is cringe. Serious people in AI don’t speak like this. It’s out of touch to the end. Important note: emails like this are hard to pull off. Many people put in many hours to write this, in an effort to fix a problem that wasn’t their fault. This isn’t a comms problem. The obfuscation, lack of clarity, and fundamental failure to grasp the problem are due to a failure of leadership. A poorly written email is just the means through which that failure is revealed.
1
6
455
None of the freakout over this in the last week would be even slightly surprising to anyone who has paid attention to how these issues have played out in video games (often in a much more subtle, nuanced, and varied way than one might imagine, too).
1
1
135
The gaming industry has a many-year head start on the generative AI industry in figuring out how to get the GPUs to show people the exact blend of counterhistorical imagery they want, and perhaps there's something for GenAI people to learn there!
1
2
141