Bertil Hatt retweeted
It was a BIG night tonight, the launch of London Cycling’s report, “What stops women cycling?”
1000 women were surveyed about the abuse they face on bikes in the capital, and the results are astounding.
9 out of 10 women have faced aggression or abuse on the road.
🧵 1/14
Bertil Hatt retweeted
💥 "GET OFF THE ROAD, BITCH!" 9/10 women cycling in London face abuse says new LCC survey. CW: swearing, abuse descriptions
Sign the #FreedomToRide petition & share to add your voice >> action.lcc.org.uk/safe-cycli…
Full report: lcc.org.uk/freedom
Bertil Hatt retweeted
We've just launched a new careers page. Our goal is to build a small, high-quality and super fast-moving team. So this area is really important to us.
plain.com/careers
We had a lot of fun making this and went beyond the average careers page – below my favourite details 🧵👇
Bertil Hatt retweeted
🚨 IMPORTANT : le très sérieux journal Washington Post, vient de publier un article très fouillé sur l'appui de la Russie au Rassemblement National de Marine Le Pen depuis 2022. Pire : Poutine a mobilisé des armées de trolls pour diviser la société française. Je vous explique 🧵
Bertil Hatt retweeted
Are there any public examples of tiered documentation?
By "tiered documentation" I mean having one set of documentation for human users and having another set of documentation specifically for LLM training.
Bertil Hatt retweeted
I was one of the many economists whose lives Bob Solow influenced.
He encouraged me to write my very first paper as a PhD student, “The productivity paradox of IT” and gave me helpful advice along the way.
Rest in peace.
Bertil Hatt retweeted
What are the best #ABTesting videos you’ve seen?
I looked for those videos that were over 10 minutes and had more than 100 thumbs-up per year since they were uploaded to YouTube.
Here’s the hand-curated list: bit.ly/bestABTestingTalks
What did I miss?
Post with details: linkedin.com/posts/ronnyk_be…
Bertil Hatt retweeted
IMO, one of the killer apps for LLMs is going to be understanding what’s happening in complex bureaucracies like large companies and governments. These are very inherently information constrained environments, but also have abundant documentation. Sorting out privacy will be key.
Bertil Hatt retweeted
If you're a smart, hard working doctoral student on the job market this year (or know one), please share this opportunity to work with me and my amazing colleagues at the Stanford Digital Economy Lab.
The Lab is hiring! Check out our postdoctoral openings in three research areas:
1. Human-AI Interaction Systems and Civil Discourse
2. Measuring the Digital Economy
3. AI and the Economics of Aging Populations
Apply now 👇
digitaleconomy.stanford.edu/…
Bertil Hatt retweeted
Did I get Sam Altman fired??
I don’t think so…
but my *full* Red Team story includes an encounter with the OpenAI Board that sheds real light on WTF just happened
I've waited a long time to share, and even the @OpenAI team will find new info here! 🧵
xitter.gq/labenz/status/16357542…
Humbled to be credited as a Red Teamer in the GPT-4 Technical Report.
I spent 2 months testing GPT-4, and I have no doubt it will change the world.
Research paper here: cdn.openai.com/papers/gpt-4.…
Bertil Hatt retweeted
System 2 Attention (is something you might need too)
paper page: huggingface.co/papers/2311.1…
Soft attention in Transformer-based Large Language Models (LLMs) is susceptible to incorporating irrelevant information from the context into its latent representations, which adversely affects next token generations. To help rectify these issues, we introduce System 2 Attention (S2A), which leverages the ability of LLMs to reason in natural language and follow instructions in order to decide what to attend to. S2A regenerates the input context to only include the relevant portions, before attending to the regenerated context to elicit the final response. In experiments, S2A outperforms standard attention-based LLMs on three tasks containing opinion or irrelevant information, QA, math word problems and longform generation, where S2A increases factuality and objectivity, and decreases sycophancy.
Bertil Hatt retweeted
New mega-paper on how to build recommenders that support human values!
- what values should a recommender support?
- what metrics can provide feedback?
- what do we know about designing for these values?
- what policy approaches might incentivize this?
dl.acm.org/doi/10.1145/36322…
Bertil Hatt retweeted
Major flood on the streets due to heavy rains in the Dubai, UAE 🇦🇪 (17.11.2023)
Source: Saudi Weather gr
TELEGRAM JOIN 👉 t.me/Disaster_News
Bertil Hatt retweeted
Awesome demo showing the power of TypeScript and @typechat to get well-typed, structured data from LLMs 🤯 from @drosenwasser
Bertil Hatt retweeted
In his new book, @GergelyOrosz mentions the need for a work log. I made an entire presentation about that!
youtube.com/watch?v=HiF83i1O…
Bertil Hatt retweeted
LLM OS. Bear with me I'm still cooking.
Specs:
- LLM: OpenAI GPT-4 Turbo 256 core (batch size) processor @ 20Hz (tok/s)
- RAM: 128Ktok
- Filesystem: Ada002
Bertil Hatt retweeted
When you say “I want to make the world a better place,” you add an implicit “I want power and should be trusted with it.”
Increasingly, I think “Copenhagen ethics,” as critically defined in Jai’s influential essay, is basically correct. When you try to solve something, you assert power over it. That matters.
Asserting power is not a neutral act. You tell people: I have identified a problem. I know what will make it better. Trust me—let me step in, let me Do Something—and the world will become a better place.
That’s a good thing, often. Almost every good work comes from an assertion of power. But by interacting with situations, you change them. You make others more or less likely to step in. You direct resources one direction or another. Should you succeed, you tend to acquire more power.
And when your good intent goes wrong, because you have asserted power, it goes wrong on a scale unimaginable for the regular, powerless person—the one who never interacted with the problems at all.
Effective Altruism is a noble concept that draws many well-meaning people. It is also, fundamentally, an assertion of power. The fall of SBF is a good illustration of its failure state: power irresponsibly gained and redirected, with countless left to pay the costs of its failure. When your circle of moral concern—those you feel a responsibility to act on behalf of—is the entire world, you assert power over the entire world.
This is at the front of my mind as I watch the tragedy unfolding in Israel and the commentary on it. Every act of public commentary is an assertion of power: Listen to me. Think and do as I say. My thoughts should impact this setting.
I get a taste of this every time I choose to cover a topic, sometimes intensely so. I learn deeply personal things about people and situations far from me. In my recent article about the backstory behind a beach altercation, I dug through half a decade of ugly and intensely personal drama between people I had never heard of prior to the altercation, and needed to choose what to attach their names to in the public eye. When I assert that power carelessly, I open myself to justifiable blame.
Jai’s post is worth reading. It’s full of compelling examples of people trying to do good in neglected areas and being lambasted for their trouble. But I think in neglecting the role of assertion of power it reflects a common blind spot among altruistically minded people, encouraging a fantasy: that you can Do Good without opening yourself up to the costs of Doing Good Wrong.
Do good. Say things worth saying. Get involved in causes that matter to you. But understand that every time you do so, you enmesh yourself in a web of responsibilities. The assertion of power is neither trivial nor costless.
Bertil Hatt retweeted
#SQL dialect madness part 2: how do you add a day to a date?
































