they are more likely to learn to act as expected in precisely those circumstances while behaving competently but unexpectedly in others. This can surface in the form of problems that Perez et al. (2022) call sycophancy, where a model answers subjective questions in a way that flatters their user’s stated beliefs, and sandbagging, where models are more likely to endorse common misconceptions when their user appears to be less educated
Below are some selections from the list (quoted verbatim):
—
1. LLMs predictably get more capable with increasing investment, even without targeted innovation
There are substantial innovations that distinguish these three models, but they are almost entirely restricted to infrastructural innovations in high-performance computing rather than model-design work that is specific to language technology.
2. Specific important behaviors in LLM tend to emerge unpredictably as a byproduct of increasing investment
They’re justifiably confident that they’ll get a variety of economically valuable new capabilities, but they can make few confident predictions about what those capabilities will be or what preparations they’ll need to make to be able to deploy them responsibly.
4. There are no reliable techniques for steering the behavior of LLMs
In particular, models can misinterpret ambiguous prompts or incentives in unreasonable ways, including in situations that appear unambiguous to humans, leading them to behave unexpectedly
6. Human performance on a task isn’t an upper bound on LLM performance
they are trained on far more data than any human sees, giving them much more information to memorize and potentially synthesize
The list got me thinking… instead of framing as “AI replaces X job”, I think the actual outcome is more like “AI recreates X job”, in much the same way that ATMs recreated the bank teller’s job, and personal computers recreated the typist’s job, and Photoshop recreated the graphic designer’s job…
Implicit in this, is that change is inevitable and outcomes will favor those who best adapt.
Just some thinking aloud…
Content creator –> after AI –> Human does more editing, curating, and aggregating (eg, across different media types)
Journalist –> AI –> Human does more primary research (developing sources, interviewing), editing
Teacher –> AI –> Human does more coaching (emotional support), planning (what to learn when), problem solving (when students are stuck)
Customer service rep –> AI –> Human does more complex issue resolution, relationship building, sales development
Social media manager –> AI –> Human does more editing and curation, community and relationship building
Translator –> AI –> Human does more fact checking, editing, research
Musician –> AI –> Human does more mixing, curating, multimedia, live performance, inventing new musical styles
Not insignificant, too, that several of the jobs on the list — such as web developer or social media manager — didn’t exist in their current form as recently as a few decades ago, and were also enabled (or transformed) by similar mega waves of technological change (eg, personal computers, smartphones, the internet).
I do think AI has surprised in the following important way: Even as recently as a year ago, most people would have assumed that the creative fields (broadly, activities like making art, writing fiction, composing music) were less at risk than the more repetitive, linear, analytical fields. Today generative art and LLMs have definitively proven otherwise.
Did some great tech internships
Cofounder Evan was TA at Brown, most brilliant person he knew
Knew he could learn a lot from Evan even if it was a failure
Got Thiel Fellow – $100K over 2 years – enabled him to focus on Figma – no dilution, helped with network
YC is good for enterprise – sell B2B, get initial customers
// Elad – early YC Demo Days – only 10-15 angels in audience – had program to help educate angels
Useful to ask “why now” when you start a company
For Figma, in 2012, WebGL arrived, initially experimented with computational photography, then went into design
Thought it would only take 1 year, but took 2 – could have moved faster if they hired faster
Microsoft told them they had to start charging so they could spread it internally (at Microsoft) – knew then they had product market fit
Customer wrote 12 page document telling them what they should go build – another moment of product market fit – the market was there, was trying to pull product out of them
Never managed before Figma, it was tough as they scaled
Bringing first manager was catalytic – meant he could learn from the manager
// Elad: Bill Gates would hire COO, learn from him, then fire him and hire another one, learn from him, etc
Interesting areas
-people move in herds a lot – right now people excited about AI
-look for different under explored areas
-lots of good ideas out there – more important is find personal passion, if you’re 3 years into an idea you hate, you’ll burn out, happened with his friends
Thought he was late to crypto, but people tell him now he was early
Got emails in 2009 talking about bitcoin
In Thiel Fellowship – people were very excited about it (first bubble in 2013-ish)
Got more interested in Ethereum’s technology
Wife started crypto company Ironfish
They talked about crypto collectibles at time (digital items fascinated him) He really liked to play NeoPets – virtual economy – felt exact same as Ethereum and NFTs Buy things you’ll want to keep forever – only sold 2 NFTs in his life
Problems in crypto
-privacy very important – holding crypto is big security risk
-scalability taking off now
-regulation desperately needed – lack of it is blocking crypto’s advancement – especially in the US, crypto will move elsewhere if we don’t solve regulation
–
Every industry will be touched by AI
Pace is staggering
Completely new tooling method Already world changing tech even if it stops improving
AGI is difficult to define – AI will make fundamental research contributions By 2030, there will be AI co-author for pure math research journal
New version of Turing test – multiplayer AI – bunch of humans and AIs
// Elad – already happened with Cicero the strategy game
How will AI impact education
AI tutors and therapists will happen – but make it local not cloud based
Colleges are scared of ChatGPT – but if it’s copying essays and hurting education, isn’t that a deeper issue? You can already hire someone to write essays
University is multiple components – mating system, credentialing system, social club
As AI proliferates, credentialing decreases in value, social club aspect increases in value
// Elad – similar to what internet did to media – mid-tier outlets got hurt, big brands thrive
Wants online universities that are better than YouTube, more structured, more social
Don’t ignore power of a well-written cold email
Communities that were on Twitter are now going private
But find those communities, learn norms, be helpful
// Elad – help open source communities
Thiel Fellowship
Haven’t seen similar programs Even Thiel applications, sometimes it’s hard to fill a class (of fellows)
Not enough people who are risk on, willing to drop out and commit
Doesn’t believe standardized testing is correlated with IQ as much as most of Silicon Valley thinks
Access to tutors, prep programs – equity component to this
// Elad – people who drop out to do startups, you gain an additional cycle of technology (eg, a few years of school, plus a few years of the first job), which is powerful experience
Still very focused on Figma
Interested in data visualization / doesn’t feel it’s done quite right
===
My friend and I started a crypto podcast called Two Degens. We talk about markets and share interesting links and sometimes invite guests. You can listen to it here.
Hi, I’ve been recording podcast episodes 1-2x a week with my cohost George (crypto OG, started WeTrust and CitaDAO), and in the latest episode we invited our friend Steven to discuss web3 gaming, state of the market, AI, and a grab bag of miscellany.
Let me know what you think. Still a long way to go to improve our comfort level and my skills as a moderator (which is totally the opposite of my ADD-ness), but we’re enjoying it and we have a lot planned!
