The discussion on AI alignment is mostly extremely naive
who or what, exactly, are we supposedly aligning it with? Californian tech bros?
As Naval points out below, humans have never been able to align with one another; that truth has been one of the principal drivers of our shared history.
And that's not going to change.
Anyone working in AI alignment should read Isaiah Berlin on his theory of value pluralism
it articulates the idea that human life is about an accommodation between mutually incompatible but equally legitimate values
for example, freedom and security. They are equally valid human goods; but they are often in tension with one another. To prioritise freedom, you often have to accept a little less security.
When we 'align' our AI with us, should we ask it to prioritise freedom or security? Who gets to decide?
Value pluralism says that many legitimate accommodations between mutually opposing values — in other words, many different but equally legitimate ways of life — are possible. There are no grounds to say that one is better than another.
So when most people talk about 'aligning' AI, they are often just talking about aligning it with one, historically contingent and context specific accommodation of values
and then...imposing that on everyone else?
That's why I believe that @EMostaque's project to make AI open source and to let a million models bloom is so crucial
there can never be 'AI alignment with humans' in the way most people are talking about; it doesn't exist.
But there can be a democratic plurality of AIs, empowering different peoples and communities to live and create according to their own value systems.
That is real alignment.
Starting a newfound leadership role? Check out the latest TechLead to see how Warren Bryant met the challenge. Or did he?
#leadership#technology#MondayMotivation
Excited to announce I’ll will be collaborating with Matt McCarty and Jerilyn Winstead on “The TechLead,” a weekly newsletter. Releases will cover #technology#leadership#management#lifestyle
What happened on 1inch in Q4?
👉1IP-11 passed an important proposal incentivizing DAO governance engagement with a new mechanism that routes a % of rev. to #1INCH stakers
👉Fusion upgrade brings fee-less txs and MEV protection
👉~250 million 1INCH token unlock
🧵on @1inch
A great report to work on as an editor. Always impressed by @kelxyz_ and @JamesTrautman_’s level of detail and professionalism as @MessariCrypto analysts. And gotta love that cover image.
1/ Despite being one of the most talked about ecosystems, the @solana ecosystem is also one of the most misunderstood.
We set out to understand the real state of the Solana ecosystem - the results of @JamesTrautman_ and I’s analysis form our:
"Solana Ecosystem Overview" 🧵
Substreams may as well be magic 🪄
Last week, the folks at @MessariCrypto processed Ethereum's entire history in 20 mins using substreams 🤯
Solana is a fast chain. With substreams & other contributions from core dev @streamingfastio, @graphprotocol can more than handle it 💪
Mundane Halloween is back!
Japan does Halloween better than anyone, but best of all is the annual "mundane Halloween" contest sponsored by website @dailyportalz.
The point is to dress up as something super ordinary.
Here are some of my favorites from this year...
Weak markets? Not for everyone! 📊
Ethereum Name Service @ensdomains held strong:
- Registrations reached an all-time-high, ⬆️ 71% QoQ
- Renewals ⬆️ 3x QoQ
How about revenue & usage?
1/n 🧵