Chris Liu

@tsarnick Some person / entity needs to be held accountable for the diagnosis and prescription. Who will that be? When will regulation catch up

Geoffrey Hinton says AI doctors who have seen 100 million patients will be much better than human doctors and able to diagnose rare conditions more accurately

@NVIDIAGeForce #BlackMythRTX Stunning graphics and amazing gameplay 🤩

Ready, Destined One? To celebrate the launch Black Myth Wukong w/ full ray tracing + DLSS 3, we're giving away an exclusive GeForce RTX 4080 SUPER, featuring stunning wrap art of Sun Wukong. Want it?! Comment #BlackMythRTX + like this post to enter!

I built a tool to make it super easy to capture thoughts, whether it’s from a walk, coffee chat, or record the entire sync. Playchat is really fast - it turns recording, organizing, and reviewing notes into one click. We’re always looking for more users! If anyone wants to discuss how Playchat can work for their team, feel free to DM me for a quick chat.

I've been using Playchat to log annoying things that happened and it'd create story that made me feel better instantly.

I really enjoy the challenges and out-of-box thinking. The game designer did an excellent job of incorporating external world through sensor, user input, and other elements to make the game more engaging. With Apple Vision Pro, I'm excited to see more real world boxes to solve!

@karpathy Is the assumption that any intermediate state is less ideal than the final copy, so the intermediate states can be used as "negative samples" in training, as in the thumb-downs in RLHF?

The ideal training data for an LLM is not what you wrote. It's the full sequence of your internal thoughts and all the individual edits while you wrote it. But you make do with what there is.

@hsinleiwang I like the analogy of having to take a driver’s license test in every city, but it’s unclear to me how multi-modality models helps mitigate this challenge. Any experts to share how transformers models can be applied to AV?

Can foundational models turn the $50+ billion investment in self-driving cars into reality? I sat in a Waymo test drive in 2019. Since then, there are many more robocars in Phoenix, San Francisco and China, but the underlying model still faces challenges in new cities. It's almost like the car has to retake the driver's license test in every new city it visits. Why is it hard? Because the model is hard to scale with one dimensional input. Recent models like Meta's Imagebind, Google's Palm E, and potentially the future of @AnthropicAI take existing models to the next level. They connect the models with visual and audio inputs, elevating it to multi-dimensional experience. It's like giving a blind and deaf self-driving car the ability to see and hear. That way, maybe the training from one city can act as a universal "driver's license" that allows the car to travel around the world.

@hsinleiwang The two examples comparing Claude and GPT-4 seem too harsh even for human. Not sure such comparisons are meaningful in the daily tasks that people intend to use LLMs for

Source: - Anthropic naming itself as Tier 1: newcomer.co/p/thinking-thr… - Berkeley’s lab evaluation: lmsys.org/blog/2023-05-1… - Research on how language model uses long text: arxiv.org/pdf/2307.03172…

Why does #Anthropic put themselves in Tier 1? I saw that Anthropic was listed as a Tier 1 AI #startup, alongside the obvious giants Google and OpenAI. Google has invented the Transformer, OpenAI has popularized ChatGPT, so it makes sense… but who is Anthropic? At that point, they haven’t released a product, but @EricNewcomer has put them in the top tier based on Anthropic's own Series C deck. While I like their product, I want to see if other AI researchers agree with them as well. I asked those who are building in the space:

@hsinleiwang Why aren’t Google LLMs on Berkeley’s leaderboard?

@Filestack As an existing customer, we've been trying to get in touch with Reza / your sales team for contract update, but haven't heard back after 2 weeks. Can you help?

@stewfortier And the quality problem comes from an underlying consistency problem -- consistent in writing something everyday, good or meh

I'm convinced that most writers don't have a distribution problem, they have a quality problem. A standout piece will almost always find an audience.

@hsinleiwang Watching @ChefRoyChoi's Intuitive Cooking class on @MasterClass

(The best) education is practically entertainment. They make you smile and feel inspired to take action. When is the last time you feel this?

Some managers create projects to match their employees' interest for retention purposes. But such projects usually become just hobbies, because they don't create values for the company. It's more effective to match employee's interest to company existing priorities. #Tweet100

MasterClass is more than an education app/website. We value transforming members’ lives. Our COO recently shared how Chris Voss’s negotiation class changed his relationship with his daughter. #tweet100

Learned a lot today by observing how my leadership delivered a tough message. #Tweet100

Any recommendations about recreating the water cooler serendipity while working remotely? #tweet100

@stewfortier AI-assisted music composition

What is the obviously-terrible business idea that you still feel compelled to start?

@CharlotCrowther And it makes a huge difference when done with genuine curiosity

Small talk is sooooo underrated.

@CatChen Agreed...

除了我之外没有其他人觉得 Bored Ape 都很样衰吗？每次看到都觉得它的艺术价值为负数。希望 Web3 多一些有欣赏价值且能获取商业成功的 NFT，这两者不应该是互斥的。

My working style used to focus heavily on getting things done. I felt so proud of myself last week when my team said "Thanks for always having our back!" I want to do more! Could you share some examples on the receiving end and/or the giving end. #tweet100