Jiquan Ngiam
@JiquanNgiam
Followers
785
Following
198
Media
38
Statuses
303
Building @Lutra_AI Previously: Google Brain, Coursera, Stanford ML Group
San Francisco Bay Area
Joined April 2012
Don't wanna be here?
Send us removal request.
Explore trending content on Musk Viewer
Trans
• 440589 Tweets
Imane Khelif
• 419518 Tweets
#EndBadGovernaceInNigeria
• 385472 Tweets
無課金おじさん
• 279695 Tweets
Coming Soon
• 221330 Tweets
#IStandWithAngelaCarini
• 193728 Tweets
#OurRockStarMarkDay
• 149455 Tweets
Kano
• 149076 Tweets
#마크생일_200만큼_축하해
• 148112 Tweets
The IOC
• 139156 Tweets
TZUYU
• 128184 Tweets
Evan Gershkovich
• 108209 Tweets
JK Rowling
• 101720 Tweets
Italiana
• 96680 Tweets
Paul Whelan
• 92914 Tweets
トランスジェンダー
• 90497 Tweets
सुप्रीम कोर्ट
• 86411 Tweets
mark lee
• 80739 Tweets
Algeria
• 76430 Tweets
HIKAKIN
• 62041 Tweets
女子ボクシング
• 60171 Tweets
ヒカキン
• 59403 Tweets
South Park
• 56334 Tweets
Argelia
• 29330 Tweets
The IBA
• 25620 Tweets
Pablo González
• 19268 Tweets
Busenaz
• 17706 Tweets
NO ONE LEAV1NG
• 14515 Tweets
ウルフアロン
• 14208 Tweets
バニーの日
• 12447 Tweets
Pinned Tweet
"When AI can do all the manual repetitive tasks for you so you can focus on truly meaningful and interesting work, it upgrades your entire work week!" -
@AndrewYang33
The impact of AI on our lives go beyond just saving time or being more productive. When we get to eliminate the
0
1
5
Gemini 1.5 is a much bigger deal than Sora.
1. The 10M context window fundamentally changes how we think about RAG. Not just large, but also effective.
2. Pro-1.5 Beats Ultra-1.0. There’s a fundamental modeling improvement.
3. It’s in production, you can use it today.
I too
19
37
297
6
1
141
@petergyang
Part of me gets excited for the tech. Part of me feels concerned that we’ll be even more glued to screens / immersive devices; will we be even further apart mentally despite being physically there.
4
0
123
1/ Excited to share more about what we’re crafting at
@Lutra_AI
! Lutra generates workflows by transforming your English instructions into code, orchestrating apps to efficiently complete tasks for you.
Our early users have used Lutra for workflows ranging from enriching an
11
15
103
Large world models by
@haoliuhl
et al. is worth checking out if you like the Gemini-1.5 and Sora results. It has 1M context window, generates videos, and is open sourced.
1. LWM shows high recall over 1M context window, and performs exceptionally well with video chat as well.
1
10
57
AI UX 2024 was a blast 🚀! Meeting
@Scobleizer
was def a highlight 🥽. We did actual live demos with tasks people came up with and it was so cool to see
@Lutra_AI
working to solve their problems 🦦
awesome event!
@swyx
@thesephist
@Mappletons
@geoffreylitt
AI Agents for regular people:
@Lutra_AI
Just one of the companies I met today that I can’t wait to get and use.
18
32
211
4
12
46
@rowancheung
We are using models to go from English instructions to fully running workflows (code). Claude3 opus has been doing great; less chatty, more precise.
@Lutra_AI
2
0
35
@rauchg
They are the only company out there that owns the entire AI stack: hardware (TPUs), research, data, product, distribution.
Finally shipping, and reacting quickly to user feedback. Bullish as well
0
0
21
@jessechenglyu
I still don’t get why I’ll want to use specialized hardware for this versus just pointing my phone at it.
Seems like a well designed app would do much better.
7
0
19
@rowancheung
You can shove in entire codebases! 1M context windows are game-changing especially for RAG-based deployments. You can just put a ton more data into the context.
1
0
15
Excited to share what we have been working on for the last few months 🚀
We are building the AI that helps build software for everyone! Lutra writes the code, interconnects with your apps, and produces AI assistants for you -- all just from English.
Introducing Lutra, a platform that allows you to easily create AI workflows. Lutra first converses with you to understand your goals, and then generates and runs code that integrates with the apps you use.
We are onboarding new users today!
Sign up at
2
2
14
1
1
14
@JeffDean
@OriolVinyalsML
Super excited to try Gemini-1.5's coding abilities, especially with context around full code bases and dependencies.
We'll be experimenting with it
@Lutra_AI
for making AI-generated workflows!
0
0
10
@jowyang
told us we'd meet hundreds for our demos at
#llamalounge
! Here's how we used
@Lutra_AI
to make a workflow that captured leads we met - including saving their information to a spreadsheet, and drafting a personalized email to them.
1. The workflow takes in Name,
0
5
10
If you’re developing agents, which framework are you using — PDCA (plan do check act) or OODA (observe orient decide act) or YOLO (just execute! no plans)?
2
1
10
Excited to see
@AnthropicAI
Claude 3 Opus out-perform
@OpenAI
GPT4Turbo in our evals and reasoning/code gen tasks. Having multiple options for models with advanced reasoning abilities is going to be a significant boon for AI applications!
0
0
9
In two weeks, we'll be showing demos of
@Lutra_AI
🦦 at one of the largest AI startup events
#LlamaLounge
🦙! We'll be showing how we built an AI assistant that can go from English instructions to fully running software that's connected to your apps. 🚀
Follow us for more
1
4
8
@rowancheung
@Lutra_AI
Yes, and for code in particular, we see less placeholders too. Working code!
0
0
7
@ericbahn
and
@dunkhippo33
are amazing! Super excited to be at camp hustle -- we'll be sharing more around the next version of the Lutra (coming soon!) and we've learned building AI products.
In October 2023, we held a small AI event for angel investors in the Bay Area. We featured a handful of our AI portfolio companies.
And Lutra AI founder Jiquan just blew everyone away.
So… what exactly does Lutra AI do?
13
15
148
2
1
3
@GroqInc
at almost 500 tokens/sec makes repeated self-reflection usable in production, changing everything.
A key technique with LLMs is to use self-reflection (Reflexion) to get the model to do better. The problem used to be that it is slow - no longer.
We can now use
2
1
7
@emollick
I’m feeling bummed that Sora overshadowed Gemini-1.5 — the improvements there were ground breaking (10M context window that was effective, 1.5-Pro Beats 1.0-Ultra indicating a fundamental improvement).
And available for access.
0
0
5
Lutra is constantly learning and improving! 🌟
If you’re curious and want to experience it firsthand, reply to here or send me a DM. We’d love to get your feedback and see how Lutra can assist you! 🚀
5
2
7
@ID_AA_Carmack
All that said, I don’t think we should discourage learning. Especially learning to code.
Calculators did not kill learning math.
0
0
7
From Prompts to Workflows! With
@Lutra_AI
, workflows are generated using just English. This video shows the end-to-end process of how to create a workflow that converts email receipts to PDFs, saving them in Google Drive.
1. Lutra first converses with you to understand the task.
0
3
6
@martin_casado
state management is hard! we've been thinking a lot along the lines of staged side effects, io monads, and how to represent real world state changes as well.
doing this right, with the right ux, will unlock a lot more user comfort with agents.
1
1
6
4/ AI-enabled programming 🤖will be 100x better, faster, and more powerful than visual programming (low/no code solutions), and allow users to scale up from small scripts to complex backends. 🚀
In the long term, AI will bring about a new way of programming – in the future,
1
1
6
We built a workflow on
@Lutra_AI
🦦 to accelerate outreach! This takes in a profile search query like "RevOps Directors" and produces a
@googlesheets
of people with their public profiles and company. Link to workflow ⬇️
You can use it to find more ideal customers to target,
1
2
5
@AravSrinivas
Love the grounded viewpoint.
The internet slurps up extreme narratives, which often feels unnecessarily polarizing to me.
1
0
4
@itstimconnors
Here's how to do this with
@Lutra_AI
- full setup under 4mins
We're on-boarding users! DM if you're interested to get early access.
0
1
5
The hard part of designing AI-native products isn't figuring out what the AI can do, instead it's figuring out what the AI cannot do, and where it fails -- and then designing the product so that users can succeed despite that.
0
0
5
2/ We developed an execution environment for AI-generated code, that is designed for security and reliability, while being connected to your apps – so that anyone, even non-technical users, can benefit.
Our journey started by understanding how best to use AI to produce quality
1
1
5
@emollick
We'll soon be able to give AI agents a screen recording example, from which they'll be able to infer what's going on, and then create automations that can take over repetitive work.
0
1
3
Deep Learning with NVIDIA GPUs started way back in 2009 -- this paper [1] from
@AndrewYNg
lab in 2009 was the first to introduce the idea. Through the last 15 years of progress with Deep Learning,
@nvidia
touched a market cap of $2T last week.
I have memories of staring at lots
0
0
4
We had a blast at Llama Lounge demo-ing Lutra -- amazing vibes, got to meet a ton of great people! Kudos to
@jowyang
for another successful Llama Lounge🦙
2
0
3
5/ Get early access to Lutra today!
Sign up here:
and follow
@Lutra_AI
for more real-time updates!
1
0
3
Claude 3 Haiku’s reported coding performance on HumanEval exceeds GPT-4 & Claude 3 Sonnet 👀 while being way cheaper, faster…
@AnthropicAI
@OpenAI
0
1
3
3/ Even better, Lutra has the ability to call AI-powered functions - this is like AI inception 🌀. So rather than customizing a prompt by hand, Lutra comes up with the right prompt based on the task you give it. It’s almost mind-boggling!
This enables Lutra to perform a wide
1
1
3
@paulg
I thought it’d be more a O(items * house_size) if you use DP. Not too different from a big knapsack problem.
1
0
3
@tunguz
@drummatick
the power of these transformer models often come from the data it’s trained on, I’d try fine tuning from an OSS checkpoint. The power of the transformer architecture is that (a) it’s easy to adapt to different problems/modalities, (b) lends itself well to broad pre
1
1
3
Generated cute kittens and puppy videos are going to take over the Internet.
🙀🐶
"cinematic trailer for a group of samoyed puppies learning to become chefs"
Video generated by Sora.
66
180
1K
1
1
3
@rauchg
@nextjs
@v0
@MistralAI
@anyscalecompute
Pretty cool! Even more if you can emulate dang’s moderation automatically with an LLM 😂
0
0
2
@GregKamradt
We're making it possible to create these kind of workflows using only language!
DM me if you'd like early access to
@Lutra_AI
0
0
3
It's pretty remarkable how good and cheap mistral medium is -- just shy of gpt4 performance, but at a fraction of the cost.
Chatbot Arena is a unique LLM benchmark that
(1) captures the elusive "in-the-wild vibe" to some degree;
(2) live updated and adaptive;
(3) relatively hard to game.
It works like democracy: people vote for their preferred models on the prompts they actually care about. Unlike
42
102
513
1
0
3
Visual programming (Zapier/Make/legacy no-code) will be superseded by AI generated code.
I have zero tech bg and can’t code, struggled learning even Zapier and Make (tho I persisted and spent weeks on the learning curve).
Lutra completely changes the game for ppl like me.
1
0
1
1
0
3
I use a version of this automated workflows to categorize my emails and it has totally accelerated how clear out my inbox everyday.
My version also pulls out marketing email unsubscribe links so that I can easily unsubscribe from all of them.
#gmailhacks
#gpt4
#gtd
0
1
3
@DrJimFan
Remarkable how good and cheaper mistral medium is -- just shy of gpt4 performance, but at a fraction of the cost.
0
2
3
@thesophiaxu
@Lutra_AI
@getlindy
one thing we do quite differently from existing agents is a clean separation of logic and data.
this helps reliability and security - mitigating prompt injections. long running workflows are also reliable!
i think it's great to have more startups working actual productive use
0
0
2
If you like AI agents, you have to try out Lutra AI - Thanks
@KaranVaidya6
for the recommendation!
Agent platform recommendation from someone who has used at least 10 of them, if not more:
After using a bunch of agentic/AI Employees platforms, I have more and more started liking
@Lutra_AI
more and more.
1. The dataset they produce is good.
2. Their dogmatic approach to using
2
2
13
0
0
2
Find the workflow at:
If you don't yet have access to
@Lutra_AI
, like this post and send a DM/reply below, and I'll send you an invite code!
2
1
4
@MatthewBerman
I’m using
@Lutra_AI
for email drafting. You could give it more context by asking it to draft based on say instructions from a Google doc, and also find recent emails from that sender
0
0
2
0
0
0
We had a blast showing demos of
@Lutra_AI
yesterday — it goes from prompts to workflows! Lutra generates and runs them for you.
Will be posting the demos here soon! Look out for them
The Herd Assembled: LLAMA LOUNGE hosted at GenLab in downtown SF. Photos by
@erickd
-Aprox 300 AI Builders and VCs
-Influencers, Press, and media attended
TEN AI AGENTS ON DEMO FLOOR:
-Agent Lunar: Digital workers built for small businesses
-AgentOps: Agent benchmarking,
17
18
162
0
0
1
One of my favorite GPT hacks is to ask it to target a reading ease score of 75.
GPT tends to produce flowery language, and this helps make the text it produces be much more readable instead.
#ChatGPT
1
0
2
@MatthewBerman
@hnshah
@zapier
This is awesome! We’re building tools at
@Lutra_AI
to make such workflows super easy to build. Happy to get you early access — DM me!
1
0
2
@Scobleizer
@Lutra_AI
@swyx
@thesephist
@Mappletons
@geoffreylitt
Awesome -- let us know what works and not, we still have lots to fix and improve!
0
0
2
@bernhardsson
Seems like we'll be in a vicious cycle here -- more code is AI written -> AI training further emphasizes AI written code -> more developers prefer to use environments that enable AI written code -> more code is AI written....
0
0
2
@dan_abramov2
Perhaps better to brand the idea as “React Component Protocol” (RCP); the client/server aspect seem to be details, but the focus is that it’s a protocol to send components over the wire (iiuc)?
0
0
1
@matei_zaharia
@lateinteraction
Compound AI systems bring models, feedback, external knowledge together; going to be much better than just single models/ single shot approaches!
1
0
2
@simonw
Ran this quickly on that image:
=== Low Detail ===
The books are partially obscured, but I'll list the titles that I can read or identify:
1. "Mapping Hacks" by Schuyler Erle, Rich Gibson, Jo Walsh
2. "Programming Hacks" (full title and authors not visible)
3. "The Return of
0
0
1
Excited to demo new capabilities of
@Lutra_AI
at the next Llama Lounge!
We're exploring the forefronts of using AI generated code to enable anyone to automate their work and life.
I host an event series for AI startups called "Llama Lounge" the next event is in Feb, focused on AI Agents. Ten AI Agents will be on the demo floor
Meet 10 AI Agents on the Demo Floor:
-Agent Lunar: Digital workers built for small businesses
-AgentOps Agent benchmarking,
6
6
50
0
1
2
More kids should learn how to program, not less!
1. Programming is about systematic thinking, not just coding.
2. Systematic thinking is important- learning how to break down problems is a critical skill.
3. AI accelerates learning: it raises the bar to be a good programmer.
1
0
2
“If you can type into a Google, you can automate part of your job.”
Love the quote! One of the best use cases of
@Lutra_AI
is to get data to where you need it — from the Internet, across your systems into where you are working from, be it a Google sheet or Slack.
If you can type into Google, you can automate part of your job.
You can build full automations by simply explaining the workflow that you want automated.
In this video,
@Lutra_AI
was able to summarize & conduct sentiment analysis of Reddit posts.
One cool thing: Lutra broke
1
1
13
2
0
3
@Lutra_AI
is getting easier to use and more capable! We've been hard at work bringing in more capabilities for it to self-reflect, self-debug, and run even faster -- so that users can just give it high level directions and it'll figure it out.
As models accelerate in
1
1
2
@petergyang
We’ve found that users of AI products tend to quickly find the boundaries of where the models reach their limits.
A lot of product work is needed to bridge the experience when they inevitably get there — something that can only be done in the wild, not through fixed datasets.
0
0
1
0
0
1
0
0
1
@AravSrinivas
: fyi -- I noticed
@perplexity_ai
got the pricing wrong here --
@google
's Gemini is (confusingly) priced by character instead of tokens.
#ChatGPT
got it right, but was much slower!
Was looking to get a pricing table to make quick comparisons and cross reference it
1
0
1
@ojoshe
Agree with this. I wonder if it helps to anthropomorphize LLMs less, and instead view them as a stochastic text calculator with varying capabilities.
“If we mainly focus on the idea of human-like agents, we tend to ignore some of the most valuable forms of intelligent
1
0
1
One intuitive way to think about how AI models (Large Language Models) can are used as an "agent" is to consider using the model to predict what actions to take, given a particular state of the world. This approach some times go by the name "Large Action Model".
For example, you
0
1
1