We are going fully remote first at Quora. Most of our employees have opted not to return to the office post-covid, I will not work out of the office, our leadership teams will not be located in the office, and all policies will orient around remote work. (1/2)
It’s so incredible that we are going to live through the creation of AGI. It will probably be the most important event in the history of the world and it will happen in our lifetimes.
We are excited to launch Poe for desktop today at . With this launch, between our existing iOS app this new web interface, we believe Poe is now the fastest and easiest way to talk to AI across iOS and all desktop platforms.
Today, we are opening up public access to a new AI product we have been building called Poe. Poe lets people ask questions, get instant answers, and have back-and-forth conversations with several AI-powered bots. (1/n)
I expect we will see continual progress from here to AGI, without any period that feels like a "winter". There are just too many obvious things left to be done. And the underlying exponential growth in compute has a very long way to go.
One reason the AI boom is being underestimated is the GPU/TPU shortage. This shortage is causing all kinds of limits on product rollouts and model training but these are not visible. Instead all we see is Nvidia spiking in price. Things will accelerate once supply meets demand.
We are excited to announce that Quora has raised $75M from Andreessen Horowitz. This funding will be used to accelerate the growth of Poe, and we expect the majority of it to be used to pay bot creators through our recently-launched creator monetization program. (thread)
Today we are launching Poe subscriptions, which will provide paying users with access to bots based on two powerful new language models: GPT-4 from OpenAI and Claude+ from Anthropic.
I'm excited to be joining the OpenAI board. I continue to think that work toward general AI (with safety in mind) is both important and underappreciated, and I'm happy to contribute to it.
Today we are starting a beta test for a new product called Poe. Short for "Platform for Open Exploration", it lets people ask questions, get instant answers, and have back-and-forth dialogue with AI.
Possibly the worst takeaway you could have from web3 being overhyped and then failing to deliver is that the optimism around AI means it will fail to deliver. If there was ever a case for first principles thinking or digging into details beating pattern matching this is it.
Today we are launching creator monetization for Poe! This program lets any bot creator on Poe generate revenue. This is a major step forward for the platform and is the first program of its kind, so we are very excited to see what it lets everyone create. (thread)
We are launching a new feature on Poe today: user-created bots. This initially allows anyone to create a new bot based on a short text prompt and an existing bot used as a base. We currently support Claude instant and ChatGPT as base bots.
We (or our artificial descendants) will look back and divide history into pre-AGI and post-AGI eras, the way we look back at prehistoric vs "modern" times today.
AGI (AI at avg human level for any job that can be done remotely) will massively impact GDP growth. We might expect >50%/year. Assuming a continuous run-up to AGI, we should expect GDP growth to start to accelerate as we get closer. When will this be noticeable in GDP metrics?
It's common in in-person meetings to "go around the room" and have everyone share their view on a topic. This is much harder over Zoom since everyone sees the group in a different order, and eye contact doesn't work. My thoughts:
There is a category of formal proofs that various things are impossible: the halting problem, the traveling salesman problem, Arrow's theorem, etc. As a general rule, I don't like these and don't think they are very useful in building or thinking about real-world systems.
Today we are adding an important new capability to Poe: multi-bot chat. This feature lets you easily chat with multiple models in a single thread. (1/n)
A cool thing you can do with the features we launched today: upload a file and then share the conversation so others can ask questions of the AI about it. For example, here's today's court decision about the FTC vs. Microsoft + Activision Blizzard:
Economists seem to consistently be the most dismissive of AI existential risk concerns, out of all groups of people who think seriously about the future. Why is this and what can we learn from it?
Finally and relatedly, we should really stop worrying about super intelligent, evil AI. I am impressed by, e.g., GPT-4, but it is NOT intelligent and will not take over the world.
It can make huge damage to inequality, wages and democracy without being super intelligent.
Amazing to see Nvidia revenue already above Intel and Samsung when we are only one year into the AI boom. There are many orders of magnitude left to go.
Poe for mobile web is launching for Android today, available at . With this launch, we believe
@poe_platform
is now the fastest and easiest way to talk to AI across all platforms, with real time sync so you can continue conversations across all devices.
This paper, a very narrow result which I'm sure only holds under a lot of assumptions, provided a good Rorschach test for people's views on AI... surprising how many of them are expecting progress to stop.
New consideration in choosing a programming language: how good are LLMs at writing it? Pushes the balance significantly toward Python and away from new languages, at least until the next step function increase in LLM power.
So weird that you gain status in the ML community by talking about how impossible AGI is, and it’s a taboo to be excited about it. Imagine if JFK or those running the Apollo program inspired their followers by saying “we won’t get to the moon anytime soon”.
Fine tuning is going to do two things:
1. Enable a much larger industry contributing to AI progress
2. Create much greater diversity in models and applications
It has been a productive process the last three months working with Bret, Larry, Sam, and Greg to improve OpenAI’s governance. I’m excited to welcome Fidji, Nicole, and Sue today and am looking forward to working with Sam and the company.
Someone is going to train a model to make professional-quality remixes of songs and it’s going to be awesome to be able to dial up the familiarity vs novelty of listening to any song you want. It will breathe new life into your relationship with all existing music.
We're excited about the Poe API and have released a repository with some basic examples on how you can use
@LangChainAI
and
@FastAPI
to build Poe handlers. Check it out here:
Today we’re introducing a new way for model developers and bot creators to generate revenue on
@poe_platform
: price per message! Creators can now set a per-message price for their bots and generate revenue every time a user messages them. Thread 👇
Introducing the Poe API v2! This uniquely lets any bot on Poe query any other bot for free and use the output as input. Developers do not have to pay for this; instead the queries are covered under the *user's* normal message allocation on Poe.
More generally, if I were running Zoom, I would create a platform for anyone to write simple web app overlays to enable all kinds of these mediated interactions. A different app might let people in a large meeting vote +1 or -1 in real time to signal support for a choice.
We’re looking to hire a small number of experienced iOS and Android engineers to work on Poe. If you’re excited about making AI accessible to millions of people around the world, and being part of a talented and hard-working team in a fast-moving environment, we’d love to talk.
What are the strongest arguments against the possibility of an outcome where strong AI is widely accessible but there is a “balance of power” between different AI systems (or humans empowered by AI) that enables the law to be enforced and otherwise maintains stability?
But people who learn them don't typically remember all these limitations, and they end up drawing the wrong conclusions about what is possible, which tends to lead to conclusions about technology that are both pessimistic and wrong.
1) A lot of people are trying to forecast the coronavirus impact on Q2 US GDP right now. I have one small piece to add to the conversation. There is a Goldman Sachs forecast from Friday that is being widely misreported as -24% from Q1 to Q2. The forecast is actually for -7%.
I bet they fix most of these in the next 6 months. Waymo is an amazing product that has made a real improvement on the quality of life in SF and I hope our governments don't strangle it.
If you are quoting pricing for an LLM, you should include speed numbers. Without any speed target it's easy for someone to crank up the batch size and lower costs but at the expense of speed.
For this purpose, I define AGI as the ability to do anything a human can do while working at a computer with internet access (e.g. excluding robotics). I have less confidence in physical machines.
In practice, there is almost always a way to make technology do what you want. It may be a lot of work, it may be difficult, but it is rarely impossible.
We are launching two new subscription options for Quora creators today. We want to make sharing knowledge more financially sustainable, which will in turn let people share more knowledge with the world. (1/n)
In my view, this dynamic, combined with the ability for AI to accelerate its own progress, makes the advent of AI less like the internet and more like the industrial revolution. It should get society onto a permanently higher GDP growth rate, as long as we manage it well.
Zoom should add a feature for a person running the meeting to organize this. It could let them choose an order for others and then guide the group through the process, with a number over each person's video to show the order, and highlighting to show whose turn it currently is.
My bet is this starts to happen within 4 years, e.g. measured US GDP growth is 3% instead of 2% and the change is largely attributed to AI. I am ignoring all the critiques about GDP being an imperfect measure of the economy here. I think it will be good enough for this purpose.
These proofs typically apply in extremely narrow cases: worst-case adversarial input, needing to optimally maximize an outcome rather than find something good enough, or other assumptions hidden in axioms needed to make the proof work.
The real world rarely fits the assumptions these proofs rely on. You can use an approximation that may not be 100% optimal but is likely good enough. You can find things that work in practice even if the theory says they may not always. You can re-structure your problem.
Today's human-limited complexity means new laws often benefit 60% of people but are worse for the other 40% and it is a huge fight to make any progress. With AI-enabled unlimited complexity, "pareto" laws can be negotiated and passed that benefit every single member of society.
We are also initially providing all Poe users with the ability to send 1 free message per day to both GPT-4 and Claude+. Other bots on Poe, including those based on ChatGPT and Claude, will continue to be free for unlimited usage.
This may sound too optimistic but you need to remember that in-person meetings are usually far from optimal. It takes a lot of work and skill to run a meeting well, and offloading this to the software could take a lot of burden off the moderator and greatly improve most meetings.
We are heavily investing in adding features and will be prioritizing the top requests from developers. You can get started by reading the documentation on our Github at .
1/ We've gotten some questions about our stance on political ads on Quora given the news. We care about trust in the knowledge shared on our platform, and given the unique incentives involved in political advertising and what is at stake, ...
1. Speed: Poe is the fastest interface to ChatGPT, and all other bots are very fast, thanks to some significant speed improvements over the last few weeks.
I’m excited to welcome Mariana Garavaglia to the Quora board. Mariana is currently Chief Business and People Officer at Relativity Space, and we are looking forward to her counsel as we continue on our path toward our mission to share and grow the world’s knowledge.
We just launched a product I am very optimistic about: Q&A in Spaces, which allows questions and answers between a bounded group of participants in a separate namespace.
This is basically a way for an open source SaaS company to block AWS and others from free-riding off their development effort by running a competing service, while still allowing other benefits of OSS. Will be interesting to see if it becomes a norm.
It is initially available on iOS, but we will be adding support for all major platforms in the next few months, along with more bots. The best way to learn about Poe is to try it at , but read on below if you’re curious about why we’re building it. (2/n)
This is a bad idea, but just to explain one reason why: if the phone is compromised, an attacker can change the vote as it's being recorded (even after the photo of the ID and video of the face match, even if the "blockchain" part is secure).
Poe will initially be available on iOS, and it will be invite only until we work out scalability, get feedback from beta testers, and address any other issues that come up. After we get through this phase, we will open up to everyone and add support for all platforms.
GPT-4 is a major advance relative to ChatGPT and is the most powerful language model available to the world today. It is particularly strong at creative writing, problem solving, and instruction following. For example, GPT-4 solves this 2023 AIME (math contest) problem correctly.
2. Claude: Poe is the only consumer product providing access to Claude, a state-of-the-art language model with significantly different strengths and weaknesses compared to ChatGPT.
We just launched a set of features for Poe focused on augmenting human input for long-context LLMs, along with Claude 2, ChatGPT-16k, and GPT-4-32k. See launch thread below:
Today we are launching new, more powerful models with increased context windows on Poe, including the just-released Claude 2 with its 100k-token window length, along with a set of new tools to help everyone take advantage of them. (thread below)
Poe currently supports ChatGPT from OpenAI (at ), Claude from Anthropic (at ), and a few other AI-powered bots. Support for many more will be coming over the next few months. A few things make Poe unique today:
@ESYudkowsky
@ylecun
What's the best summary of your logic for why AGI implies near-certain extinction of humans? I'm not sure this has been laid out very clearly for people.
We are running a new programming contest featuring some problems inspired by challenges we've faced in building our product. It's accessible worldwide via two 4-hour divisions, 12 hours apart from each other.