Today, youChat goes live. Open, broadly capable, conversational AI for search with knowledge of recent events and citations of sources. Search and chat of the future: you.com/search?q=what+was+th…

It's always amazing to see your past interns and students lead multi-billion dollar companies. He's one of at least 4 of mine that I can think of. HuggingFace and Commure are two others. <3

Introducing YouCode: The best search engine for developers. Available now at code.you.com. A 🧵 with our features that make you more efficient when coding. We’d love your feedback to make it better for all kinds of programming, from frontend, web3 to backend and AI.

Bored by generic chatbot answers? Today, @YouSearchEngine is moving closer to our vision of being your personal AI assistant for search. Introducing Smart Personalization ✨ No need to type up your preferences, it learns w/ transparency It's about understanding and helping you

Today, we launch YouAgent, an AI agent with code execution for more accurate answers to complex math and science questions. Unlike other consumer chat engines that cannot do reliable, multiple-step reasoning, it will also compute your mortgage accurately ;) A🧵 >

Introducing You⚡Pro One AI chatbot to search the web, chat, code, write and create images. Now supercharged with GPT-4 and Stable Diffusion XL. Stay up-to-date with the newest AI models, tools, and advances - while paying less.

Excited to introduce the AI Economist: Extends ideas from Reinforcement Learning for tackling inequality through learned tax policy design. The framework optimizes productivity and equality. Blog: blog.einstein.ai/the-ai-econ… Paper: arxiv.org/abs/2004.13332 Q&A: salesforce.com/company/news-…

I will be starting a new company -- with a vision I have been passionate about for a long time.

We release the largest publicly available language model: CTRL has 1.6B parameters and can be guided by control codes for style, content, and task-specific behavior. Incredible generations! Paper einstein.ai/presentations/ct… Github github.com/salesforce/ctrl Blog blog.einstein.ai/introducing…

Excited to introduce YouPro for Education—your AI study buddy. Access unlimited AI chat + search, unlimited AI writing generations, unlimited AI art generations, supercharged with GPT-4 and Stable Diffusion XL at just $6.99/month for students and teachers.

Very excited to announce the natural language decathlon benchmark and the first single joint deep learning model to do well on ten different nlp tasks including question answering, translation, summarization, sentiment analysis, ++ einstein.ai/research/the-nat…

So excited to announce our 25m Series A to build a better and more open search engine. We will focus on * youCode - the search engine for developers and * opening up the platform for more search apps, search APIs and more. A short 🧵 with more info->

Do you like prompt engineering or sharing amazing #chatGPT - like transcripts with friends? We just launched sharing of full transcripts on youChat. The best part - you can continue the chat :) you.com/search?q=where+to+wa…

If each step of an ai agent is 95% accurate. None of the 30 step work flows will work. Going from 95-> 99.9 is a similar last mile problem as with self driving cars. Easy to hack up a prototype. Hard to make it work reliably at scale.

Want your chatbot to be up-to-date? Our new YOU API empowers your LLM with real-time web search, much longer snippets and better fact recall than any other search API. We also have pure news endpoints and complete retrieval-augmented generation (RAG). more details and links >>

If you like #chatGPT -like capabilities and want to use conversational AI as your default search engine, try you.com/chat - it is free, has access to recent sources and often gives citations. You can set is at as your default via their Chrome extension.

If you studied algorithms, I'm sure you've heard of Dijkstra’s algorithm to find the shortest paths between nodes in a weighted graph. Super useful in scenarios such as road networks, where it can determine the shortest route from a starting point to various destinations. It's been the most optimal algorithm since 1956! Until now. The O(E + V log V) complexity just went down to O(E log^(2/3) V) for sparse graphs. It would be amazing if this kind of breakthrough came through AI that can code but I guess we're not there yet..

With this much excitement and energy in the LLM space, I am fairly confident in this prediction: Before the end of this year, there will be an open source model that is equivalent in quality to #GPT4 Then, AI researchers and the AI community will go to town and improve it.

It is kind of shocking how much current computing architectures (graphics processing units) define neural network architectures. GPUs made models like Transformers fast and hence popular, not neuroscience or theory. Our model space is restricted to what's fast on GPUs.

It's fascinating that some folks who are probably great science fiction writers are somehow considering their writing about AI risk as scientific research. A thread on imagined existential risks, creativity & a lack of it, history, LLMs, safety, regulation and anti-hype hypers.

We are now offering the first full web index API for LLMs. It already works better than Bing & Google SERP APIs for making LLMs more factual, up-to-date and offer citations. It's early days but you can find more details at: documentation.you.com/ blog.langchain.dev/you-com-x…

Introducing multi-modal youChat! Our chat AI is the first to show and use the content of an app when that provides a better answer than just text alone. This is a huge step forward for LLMs in search. you.com/search?q=what+does+c…

you.com Let's build a new internet together.

Introducing the Social Tab on you.com: search 9+ social media apps at once. A 🧵 on features that let you find people, places, ideas, travel inspiration & how-tos fast across: 🐙 Reddit 📷 Instagram 🐦 Twitter 🎥 TikTok 🇶 Quora 👨‍💼 Linkedin 🌅 Pinterest

Presenting: The AI Economist This is one of the most impactful lines of AI research I've ever worked on. Its implications span from immediately impactful to highly philosophical. Blog: blog.salesforceairesearch.co… Paper: science.org/doi/10.1126/scia… A 🧵 with high level take-aways:

With Google's Gemini coming out, there's pressure for every LLM to have a live web connection. Ping us at api@you.com if you want to get rid of hallucinations, keep your LLM answers up-to-date or offer citations for facts. We can help with a complete solution.

Most cost efficient way to try out #GPT4 is you . com right now. Plus it's connected to the internet and up-to-date, has the most accurate citations for facts, personalizes your answers and executes code for more complex questions. Starting tomorrow you can even try it for free!

Is there a more useful search engine out there with cooler AI features than you.com ? I think not (: Now with a beautiful AI image generation app: you.com/search?q=digital+pai…

It's nice they covered my thoughts. I learned that the journalist who writes the article is usually not in charge of its title. That's done by their editor who only thinks about clicks 🤷‍♂️ By the way. The time it takes depends on your definition of agi.

Now that openai has caught up to our old research agent from over a year ago we will soon launch our advanced research and intelligence agent which will be 10x better than what we have had and openai just launched. Give us another week or so. We're cooking 🔥🤘

YouChat just got faster and can produce even longer outputs. It is now on par with #ChatGPT plus more: Ask it anything to learn or create code, images, poems, etc with reference links and factual apps. eg where to watch movies and find similar ones: you.com/search?q=where+to+wa… 1/n

Excited to announce that I'm now an adjunct professor at the #Stanford computer science department. 🙃 Still chief scientist at #Salesforce

Growing up in Germany you're taught that nuclear energy is terrible. Yet, all my scientific friends agree that it is by far the cleanest form of energy - it is SO much better for the planet than coal/fossil fuels... I wonder if the green party/politicians can course correct?

Now that the world has access to a GPT4 level model completely open source, we will see that the fear mongering AI p(doom)ers were wrong again about the supposedly existential risk of these models.

We’re excited to introduce ARI (Advanced Research & Insights) - the first professional-grade deep research agent purpose-built for business. Instead of spending $100K+ on whitepapers and analyses that take weeks, you can now get polished PDF reports from ARI in under 5 minutes. We're opening up early access today. 🔗 below

We built you.com so you can search less and do more. To take back control of your information diet and privacy. Here are 7 tips to get the most out of our search engine. 🧵👇

A new reinforcement and deep learning model for summarization.It generates long language sequences that make sense. metamind.io/research/your-tl…

I think the open source community needs to start incorporating the concepts of * open source training data * open source human feedback collection * open source weights With LLMs it will soon be less about the code than the training data.

Personal update: After 4+ amazing years, I will leave my chief scientist role at Salesforce and transition to be an advisor. Proud of what we have accomplished with the Einstein teams: many AI products (chatbots,call coaching,+) and published research(AI Economist,CTRL,decaNLP,+)

So excited to launch our developer program to build apps on about.you.com/developers/ I think it's so important for the next search engine to be much more open, let everyone collaborate, help users actually get stuff done and let them decide what's a good result.

Rather than spending a month figuring out an unsupervised machine learning problem, just label some data for a week and train a classifier.

The calm before the storm. This feels like a once in a life time shot. The probability for everything to align was tiny. The result? A gorgeous sleeping leopard in an acacia tree with horizontal lightning on the horizon. In Masai Mara national park. #Kenya #wildlifephotography

I think one of the problems of academic publishing/peer reviews is that there is zero downside of rejecting good ideas. The reviewers of the DecaNLP paper - which introduced prompt engineering for multitask NLP with large nnets - were so wrong & held back the field for years.

Summarization is one of the most important & least solved tasks in #NLProc Problem with all #DeepLearning models: they are not optimized for factual correctness We introduce a new task, dataset and model. Work by @iam_wkr @BMarcusMcCann @CaimingXiong Paper arxiv.org/abs/1910.12840

One of the hardest problems in #AI is common sense reasoning. This paper by @nazneenrajani, @BMarcusMcCann, @CaimingXiong+I makes huge progress on this. Powerful, simple and unsupervised method: arxiv.org/abs/1906.02361 Github: github.com/salesforce/cos-e Blog: blog.einstein.ai/leveraging-…

Just 6 years ago. This NLP reviewer was 100% certain that prompt engineering to unify all NLP problems in a single neural network and just ask it any question was completely misguided and rejected the paper. It would then be cited by the GPT2 and 3 papers. Thanks @arxiv!

This Monday I had the opportunity to advice the German chancellor and his cabinet on AI, its countless opportunities as well as risks and intuitions for GPT-like models.

A new #deeplearning model for improving abstractive summarization by actually creating novel phrases. Reinforcement learning in #naturallanguageprocessing. Work by amazing intern @iam_wkr and Salesforce Researchers Romain Paulus and @CaimingXiong. arxiv.org/pdf/1808.07913.pdf

Happy to dive into more details here but the main reasons why LLMs are not making amazing discoveries are: 1) passive vs active knowledge. Like when you can understand a language but you can't put together correct sentences yourself. LLMs have knowledge passively stored but have not been trained to explore the latent space because 2) objective functions are not aligned to discovery. predicting the next word, being a good conversation partner after RLHF, finding reasoning paths to achieve a goal are not going to lead by themselves to truly out of the box thinking. 3) one LLM, like one human have a hard time making amazing discoveries. Research is best done with a swarm like intelligence so multiple agents (with either different weights or at least different prompts that push them into different states) are going to be helpful. 4) High level corporate incentives are not set up to push AI to just try things. Serendipity, even in human research, is rare and requires very long exploration paths. So the reward will be very hard to backprop through insanely long discovery DAGs. Eg, some bio researcher wants to study amphibians of Mexico, discovers that Axolotls can regrow limbs and their neoteny (retaining juvenile features into adulthood). Maybe this will result in some longevity breakthroughs but I can guarantee you that right now even the richest country in the world and its NIH could not get a study of Mexican amphibians funded because it's too far from an application many folks understand. So even for humans the paths are often too long from basic research to applied breakthrough.

Three road blocks to solve for artifical general intelligence (#AGI) 1) Massive Multitask learning with a single joint model 2) Ability of algorithms to update their objective functions in continuous learning 3) Learnable combination of fuzzy/fluid and symbolic reasoning

This strengthens my hypothesis that there is a large set of equivalent neural architectures of which the original transformer is just one sample.

The problems in a domain that you can simulate, can be solved by AI since you can create unlimited training data. This is true for games (think go, dota, etc), simple physical simulations, programming/coding environments and eventually more complex biological systems.

I've worked on academic deep learning and summarization for years. Summarization is a foundational technology for the information age and a remedy for the attention economy. Here's a🧵 for how we think and apply summarization at @YouSearchEngine

Wow. This is huge news for YOU. You.com will be a default browser option for EU countries on iOS 17.4 We're the only chat-native browser option that made the cut. Read more: 9to5mac.com/2024/01/26/apple…

Hi @elonmusk, I would believe your concerns more if Tesla didn't download powerful AI into every car. Why over-regulate some AI models instead of real applications? Language models can't say or choose what they "want." Just like a FSD Tesla can't decide where to drive by itself.

Hierarchical and interpretable multi-task reinforcement learning. This tackles some major limitations in RL: Training agents to solve complex problems that require multiple subtasks and have them explain themselves Blog: einstein.ai/research/hierarc… Paper: einstein.ai/static/images/pa…

In a surprising twist/pivot, OpenAI launches a waitlist for a neat enterprise software API. @Huggingface becomes an open ai company with amazing open source nlp toolkit. I'm loving the progress in nlp this last decade.

Let's build an #AIEconomist for the real world together! The framework for building RL-friendly economic simulations is now open source! This has so much potential for good. github.com/salesforce/ai-eco… blog.einstein.ai/the-ai-econ… blog.einstein.ai/the-ai-econ…

Fun surprising science fact for your Saturday morning. Liquids don't splash when hitting a hard surface in a vacuum. Somehow nobody really knew that so researchers actually wrote a paper. Links below. OK. Now back to working on AI :)

A reinforcement learning agent that learns to program new neural network architectures. Same/better results as LSTMs but with funky nonlinearities (sine, SeLus, etc) and new connections that result in different activation patterns😯 einstein.ai/research/domain-… arxiv.org/abs/1712.07316

Today, we launched you.com, a private search engine that summarizes the web for you 🚀. I'll be answering questions throughout the day on @ProductHunt. Thank you for your support and feedback 🙏producthunt.com/posts/you-co…

Humanity just unlocked nano-surgery: we can now inject magnetic carbon nanotubes into a (mouse) brain. They're engineered to ONLY bind to cancer cells. Then you turn on a magnetic field that makes them spin and destroy only those cancer cells. 1/

This kind of blows my mind: Imagine YOU are a data scientist, get a data file and need to build a classifier... Just tell your AI assistant to "classify this" and ... it's all DONE. Even checks for class imbalance.

Here's my mental model for when a generative ai application makes sense: GenAI is amazing when it would take a long time to create an artifact but very little time to verify its correctness. Examples: 1/

you vs pplx substance vs hype

It is fascinating how the @nytimes can report on Google MAYBE adding "novel" features IN THE FUTURE without mentioning that @YouSearchEngine launched several of these features publicly for everybody to use ~4 months ago. Starting to question my subscription and their reporting..

Today I'm excited for another step from a search engine to a do engine: you.com/write, an AI-powered writing assistant. If you are a copy writer, marketer, student, emailer, or somehow stuck on a blank page, this will make your life easier 🤯 you.com/search?q=how%20to%20… A🧵

Today we will answer questions on producthunt.com/posts/youcod… about YouCode, a developer search engine from @YouSearchEngine. Code and learn up to 5x faster 🧑‍💻 ⚡ Search 20+ apps for coding efficiency without opening a bunch of tabs 🤖 AI generated code 👉 you.com/code

Does the graphics community have any ethics folks worrying about disinformation or is that more of an AI community thing? (Honest question, because I never hear about computer graphics ethicists.)

We’re officially a YOUnicorn! 🦄 Excited to share that @youdotcom just raised $100M Series C at a $1.5B valuation, led by @CoxEnterprises. We’ve been heads down building the search infrastructure for the AI and agent future. Soon there will be more AI agents using the web than humans, but today's search wasn't built for this. Agents need deep, contextual information from both public web and internal private data to make real decisions. Our web search API delivers the most up-to-date, accurate, and fastest search results for LLMs and agents. Real benchmarks show we consistently outperform the competition on accuracy and speed while staying cost-effective. Today, we’re answering over 1 billion queries every month across enterprises like DuckDuckGo, Windsurf, Harvey, and many others. There is no other startup of recent years at this scale. Grateful to our wonderful employees, customers, partners, and investors who believe in our vision. The agentic era is here - let's build it together. Try our LLM search APIs.

The power of staying in one company, learning, improving, creating. A lot of young employees seem to think there's a need to jump around companies every few years and while you can maybe get a small salary bump each time, it's hard to create a deep foundation that way.

25 years ago, knowing "how to google" gave you an edge when it came to being more productive. Today, it's about prompt engineering and creating AI Agents on platforms like @youdotcom. That gets even more powerful when collaborating. This marks the beginning of our next chapter: The AI Productivity Engine We aren't a search engine for links. We're a productivity engine helping knowledge workers accomplish more. Our AI Agents research, give answers, solve problems, write and run code, create content, and more. We do this with a relentless focus on accuracy. We've built a model-agnostic AI Operating System, making any Large Language Model more accurate and trustworthy. Live web access, advanced search capabilities, dynamic prompting, and advanced citation logic ensure reliable, up-to-date information every time. It's a step function for productivity. Millions of knowledge workers already use us. Over 1B queries served since launch. 500% ARR growth since January 2024 We also just closed a $50M Series B today led by @Georgian_io, with @SalesforceVC, @Nvidia, SBVA, @DuckDuckGo, @DayOneVC, and others, bringing our total funding to $99 million. Better, better. Never done.

First fully parallel machine translation model.No RNNs😳Less latency🙃 einstein.ai/research/non-aut… einstein.ai/static/images/pa… cnbc.com/amp/2017/11/06/sale…

Do you want to help build the next defining internet consumer company on values like trust, facts, kindness and equality? We're hiring our first team members for product, UI, UX, full stack and AI engineering. Feel free to ping me with questions :) su-sea.github.io/jobs/

I struggle with this sometimes when early career folks ask me for advice and how they could be successful ... I ruined my health for a good decade working crazy hours but feel bad about suggesting this... Maybe you can do it without that... But I don't think I could have.

and this is why the front seats are not the best place to be in the worst case scenario :||

This was possibly my most epic flying moment. So much beauty but also fear. Hope you'll enjoy it as much as I did (wait a few seconds for the big reveal). #iceland #paramotor #waterfall #adventure If you want to see more of this: instagram.com/richard_explor…

Keep AI open.

Best Paper Award at @ACL2019_Italy! The paper improves dialogue systems through transfer and few-shot learning. Congrats to our co-authors and our intern and future member of the @SFResearch team @jasonwu0731 Paper link: arxiv.org/abs/1905.08743 Code link: github.com/jasonwu0731/trade…

Can you overfit on your data? —> Collect more training data. Do you underfit your data? —> Train a bigger model. It is surprising how many neat research results came from this simple formula.

My TEDx talk on the present and future of artificial intelligence is online piped.video/8cmx7V4oIR8

I'm happy for Tomas Mikolov and his co-authors to have won the test of time award. It's well deserved and I congratulated him on Facebook already a few days ago. He is Schmidhuber-ing a little but it's understandable. As is often the case in academia: Success has a lot of parents, only failure is an orphan. Often, intelligent and creative minds see connections and isomorphisms between ideas and models. Once a topic has excited smart and creative minds they tend to have similar ideas. Many ideas are sort of "in the air." It's also true that a lot of ideas have been mentioned at some point but not implemented or executed on at scale. I try not to attribute malice where ignorance can suffice. When I first started playing around with nnets in 2009 I tried to merge ideas from standard neural nets (that were all on fixed length inputs, images, sound snippets, fixed text windows) with those of NLP (like grammar). I was just excited, exploring and had this idea to merge word vectors one pair at a time in a syntactic tree structure. Then I had the idea that I should call those recursive neural nets since I apply the same nnet composition function on its own outputs inside a DAG... a few days later I realized I should google that term "Recursive Neural Network" (now I'd use YOU . com of course ;) and was bummed that the same idea was already invented in the 1980s with RAAMs. So just by playing around with the ideas I reinvented the wheel. At the same time, those old RAAMs didn't work at all, they were applied to binary vectors over vocabularies of ~32 words or so. There was no proper SGD, no dynamic tree structures, no large data, no useful objective functions on top, etc. Now, I did cite the old work of course but it wasn't really influential for me at all. I had grown attached to the idea way more by having invented it myself from first principles. When Mikolov writes in his somewhat disgruntled fashion that we "copied many tricks", it's a bit unclear why since we actually cited him 7 times in the GloVe paper. I also think that when you do a linear computation of embeddings, it's still a clever idea to do it over the statistics rather than for each sample individually 🤷 Among many other things, it made it easier to scale. Scaling is still under-appreciated in academia. Less so thanks to OpenAI but still. I often think transformers are just easier to scale than LSTMs but one could achieve similar results with both. Having conviction that the effort of scaling will bear fruits has alpha in itself. One might argue that scaling the hardware + data + network + engineering + teams + processes is just as important as finding a more scalable architecture for overall progress. In 2010, there were only a few folks really focused and actively working on nnets for language: Ronan Collobert, Jason Weston, Tomas Mikolov, Yoshua Bengio, myself, Chris Manning and a handful others. The field moves so fast now that unless you keep doing amazing work the new generation will quickly forget. The fields of AI and deep NLP have expanded so much in the last year and many folks who are just now joining or noticing it think "it came out of nowhere". Understandably, that upsets some folks who have been at it for a while and laid the groundwork. But hey, we should be mostly stoked that our ideas are scaling at this massive rate :) Since language is inextricably connected to thought, there's still so much more to be explored. When we combine NLP with exploration, simulation, search and tool usage, we will be able to do even more than what's possible if we rely only on predicting the next token in human text. Let's keep accelerating and building.

Prediction: Any AI problem that you can simulate and sample endlessly many training samples for can be solved with today's algorithms such as deep and reinforcement learning.

Searching for code snippets and questions can be frustrating and time-consuming. Until now: you.com will help you solve your coding questions faster and summarize the relevant web for you. Check out our newest #AI feature of code completion and join our beta :)

It's kind of absurd to insinuate that open source AI is more dangerous than any other kind of AI. The more you can study something, the more you can understand it and think of counter-measures. There's a reason most secure servers on the internet run on Linux, an open source OS.

I was honored to be invited to lead a session on AI and Europe with @Bundeskanzler Olaf Scholz and the French President @EmmanuelMacron and both of their cabinets. We covered a lot of important facets of AI, open source, research, AGI, impact on jobs, healthcare, defense etc.

YouImagine now has a MidJourney option and boy - it's faster AND creates a broader range of styles: photorealistic/paintings/drawings/etc. you.com/search?tbm=youchat&q… #ai #generativeart #GenerativeAI

Today, we are announcing the opening of our search platform. With this, we hope to tackle two problems we see with Google: 1. users do not have control over what they see, and 2. developers/organizations cannot provide useful capabilities to the search results page ->Thread

decanlp.com/ website is up! Slides motivating true multitask learning in AI and NLP from a recent talk:

Can smaller AI companies still beat OpenAI? Yes, if they focus: ARI (our Advanced Research & Insights agent) just beat OpenAI's Deep Research. By a large margin and on two benchmarks. Today, we're also introducing ARI Enterprise, giving financial analysts, consultants and knowledge workers the most accurate answers to deep research prompts. Links and details below.

This year companies will realize that an LLM by itself, even when fine tuned, will not be accurate or up-to-date and know about the world in a verifiable way. That's why we're excited to have the most accurate APIs that are always on top of recent world knowledge! api. you. com

Proposed Theorem: it’s impossible to acquire super humanity skills when relying purely on data created by humanity. Why do I say super humanity and not super human? Because for example a translation algorithm is already better than any single human in terms of how many languages it can translate between. But it’s unclear you can get super humanity language skills if you train on human language. A language beyond human understanding could include massive parallelism that humans could not comprehend as our language processing is pretty serial. But it's otherwise unclear what super humanity language skills even mean since language is a human construct and invention. That's why we see many simpler benchmarks flatten to an s curve just around human annotator skills now. The connection between complex thought and language is also an interesting question in this regard. Humanity is partially so intelligent collectively because we have different kinds of people. Some like staying in their home town and enjoy daily life as it is and has been. Some venture out in the literal or figurative darkness and Want to explore. What that means is that we need better simulations and more creative exploration of a search space and potentially multiple agents that require interaction to solve goals they agreed on together.

Introducing ProGen, a large language model trained on 280 million protein sequences that can generate viable proteins based on user specifications. A step towards AI & #nlpproc helping cure disease and clean our planet. Paper: biorxiv.org/content/10.1101/… Blog: blog.einstein.ai/progen

Hypothesis: any general function approximator that is powerful enough (eg has enough parameters) and can be efficiently trained will eventually perform similarly. The underlying substrate (LSTM, CNNs, Transformers etc) won’t matter as much as the data and objective functions.

I love LLMs but to be a viable search engine, you gotta offer more than just text. We've incorporated a lot of our apps into you.com/chat now in order to be more factual and display data in the best way. Next week we will improve the design of our citations & links

My keynote about the future of natural language process, #deeplearning, multitask learning and decanlp.com For the remote conference AI with the best piped.video/MENYCdm1eis

I haven't been this excited about an announcement since we launched youChat last year (when we were the first to connect an LLM to the internet). LLMs have enabled new ways of learning on the internet. They can give longer form, useful answers to many different types of questions. However, they also have several serious shortcomings: 1. They cannot be trained frequently enough to stay up-to-date or provide accurate citations for their knowledge 2. They hallucinate answers for important questions such as stock prices and 3. they cannot reliably reason about math, science and logic. We've tackled the first two over the last 12 months (and have seen many fast followers since - both big and small). This week we will announce a solution to the 3rd shortcoming. Stay tuned :)

Some predictions for AI in 2024: We're going to see open source models catch up to GPT4 and companies eventually using LLM operating systems that help you get these models production ready. AI-generated videos get longer and eventually really solve temporal consistency of multiple characters. LLMs will graduate to becoming agents with more powerful tools, search, APIs, coding abilities, clicking on the web, etc. In particular, AI assistants for search are helpful enough and will replace Google for many young people, students and knowledge workers. We're going to get AI music generation. The first vibe album may get released where an artist gives a general sequence of vibes. The exact lyrics will be personalized. Before that, we'll have more artists use and collaborate with AI to create new songs. I think we're going to start seeing the first rays of massive progress in bio AI. AI for biology will lead to much improved protein engineering and finding new uses for existing medicines. Not an exhaustive list, 2020s will be an exciting decade!