Henry Mao @Calclavia Twitter profile | Pikagi

Pikagi

Henry Mao

@Calclavia

4,735

Followers

467

Following

530

Media

5,695

Statuses

Ushering in the next era of AI. prev. co-founded ($6M+ ARR, exited)

Singapore

https://t.co/AlY6EJIzIN

Joined May 2009

Don't wanna be here? Send us removal request.

Pinned Tweet

@Calclavia

Henry Mao

18 days

The story of how one Chipotle burrito 🌯 led me to co-found a $6M+ ARR AI startup: I started my freshman year and met @Davidjpark96 because we were both looking for roommates. We met up at a Chipotle near campus. We made our order (no guac), sat down, and talked about our

Tweet media one

14

4

82

Last Seen Profiles

@prahjurite

@urbanbrew

@prahjurite

@thoh1585

@bokep_indo17

@bokeplokalmalam

@stw46

@eromzyyy

@WolfEyeGames

@Lifebeat

@bokeplokalmalam

@fdleon_1

@giveyouhomework

@samratrld

@bokeplokalmalam

@Chamberpaint

@jeepcherokee__

@jandakembangstw

@stwtutik

@bokeplokalmalam

@prahjurite

@BinorRaja

@stw46

@penyukastw21

@RShaverSports

@QuentenS25455

@SARC8427

@Gluhen03

@bokeplokalmalam

@MuthuriRickie

@jandakembangstw

@JuvenalYar59284

@luv2trap

@VideoUkhty

@sinihsamatante

@BinorRaja

@Calclavia

Henry Mao

1 year

4 years of building. 3 years of no traction. 2 years of moving to Malaysia to reduce burn. 1 year of insane growth. Now at $200K MRR!

Tweet media one

89

31

2K

@Calclavia

Henry Mao

8 months

@karpathy Taken to the limit, AI coding is perhaps more akin to an advanced compiler. Compilers take high level code and turn it into low level instructions. AI compilers do the same but take natural language or human goals as input.

5

8

139

@Calclavia

Henry Mao

1 year

Watched @karpathy 's talk on the State of GPT. The most interesting slide was how LLMs learn a spectrum of performances, and you need to ask LLMs for good performance. Feels similar to Upside down RL () where you sample trajectories from desired rewards.

Tweet media one

1

3

81

@Calclavia

Henry Mao

3 months

@johnloeber Elo rating is better for so many other things: - hiring candidates - sprint task point estimates Things are more often relative rather than absolute.

0

0

62

@Calclavia

Henry Mao

8 months

@svpino The missing key here is the non-linearity (e.g., ReLU). Without non-linearity deep learning is just linear regression. With just a tiny bit of non-linearity, you can approximate many things.

2

1

54

@Calclavia

Henry Mao

1 year

We've just been accepted to @vercel 's AI accelerator! At @whoisjenniai , we've been using Vercel since day one (~4 years ago) to ship our Next.js front-end back when it was called Zeit. Great to see how fast Vercel has grown and how it's supporting the AI builder community.

Tweet media one

8

0

54

@Calclavia

Henry Mao

1 year

@Rafael_L_Spring 1. Making the features as simple as possible with good UX. Focused quality is better than quantity. 2. Serving a market that's unpopular

4

1

45

@Calclavia

Henry Mao

1 year

@karpathy This shows us there are a lot of opportunities to potentially self-train from the model's own data. For example, GPT could self-train to be more consistent by sampling A -> B, B -> A, and asking the model to resolve the conflict, then using that as new training data.

3

0

42

@Calclavia

Henry Mao

8 years

Really excited to say that I'll be working at Blizzard over the summer on Overwatch! :D

8

2

36

@Calclavia

Henry Mao

4 months

@_Felipe That's where Protobuf shines!

2

0

36

@Calclavia

Henry Mao

1 year

@sshkhr16 Similar reason why I dropped out of my PhD program in 2019. Industry seems to be leading and there are too many opportunities. Academia can't compete with industry budgets, especially when training those giant models!

3

1

32

@Calclavia

Henry Mao

3 months

I tried 4 different VS Code plugins for AI chat: - @SourcegraphCody - @Cursor - @CodiumAI - @GitHubCopilot So far, Cody's chat has the slickest UX, and you can use Claude Sonnet 3.5 as the underlying model, which is way better than Github Copilot's chat model.

3

6

30

@Calclavia

Henry Mao

1 year

Introducing narrative citation generation in Jenni. Jenni @whoisjenniai can seamlessly weave in citations in your writing in a more natural way instead of only adding them at the end. Took a lot of work to get this working. Coming very soon!

4

4

27

@Calclavia

Henry Mao

1 year

@wangzjeff At least VS Code never ghosts. 😂

2

0

26

@Calclavia

Henry Mao

7 months

2019: we started Jenni in our college basement fine-tuning GPT-2 2020: all our pilot users churned due to COVID 2021: we got rejected by YC W21 2022: we found PMF after many pivots 2023: we reached $100K MRR Now passing $400K MRR (2x in 5 months)!

Tweet media one

3

0

30

@Calclavia

Henry Mao

1 year

Had a packed month visiting SF and meeting all types of AI founders/startups! The AI software tool stack is being defined this year: - Vector databases are solving the limited context memory of LLMs, like @trychroma and @pinecone - Data source connectors like @carbon__ai are

5

1

23

@Calclavia

Henry Mao

1 year

Happy to announce that our startup @whoisjenniai will be joining & funded by AIGrant ( @natfriedman @danielgross @Hersh_Desai )! I'll be flying back to San Fransisco to attend the summit at the end of Sept. Let me know if you want to grab a coffee in person!

@Hersh_Desai

Hersh Desai

1 year

Jenni is an AI workspace that helps researchers publish their next groundbreaking paper.

2

3

20

1

1

21

@Calclavia

Henry Mao

1 year

Been using @weights_biases since 2018 in my PhD program. Back then it was an obscure experiment tracking tool that helped collaboration between people in my lab. Now they've got a full conference with many attendees. Crazy how fast AI has exploded in the last few years.

@Calclavia

Henry Mao

1 year

At @weights_biases fully connected conference!

Tweet media one

0

1

8

0

5

20

@Calclavia

Henry Mao

2 years

@frantzfries I'm running an LLM based startup @whoisjenniai . While ChatGPT is great general purpose tool, there are many user needs that are not covered. Focusing on those pains is key. Just because there's Google search doesn't meant there's no need for Wolfram Alpha.

0

0

21

@Calclavia

Henry Mao

3 months

Had a blast hosting Singapore's first AI Tinkerers meetup! Shoutout to the awesome demos we saw last night: @hellogabbo : run local open-source models with Jan & Cortex Jia Qi: speech separation @erniesg : RAGgaeton, Scaling AI-Augmented Writing for Any Content Type @jetnew_sg :

Tweet media one

Tweet media two

2

6

22

@Calclavia

Henry Mao

11 months

@pmddomingos Fake artificial intelligence = real intelligence?

3

0

18

@Calclavia

Henry Mao

9 years

Looks like I won't be leaving Minecraft soon. Officially got a job that's related to MC Modding. :)

3

1

17

@Calclavia

Henry Mao

8 months

Why AI copilots are better than AI agents: As we scaled Jenni to 250K+ monthly active users, we tested many designs for our AI features. Some designs help users complete tasks autonomously while others assist users via human-in-the-loop. We found that the copilot experience

Tweet media one

@cognition_labs

Cognition

@cognition_labs

8 months

Today we're excited to introduce Devin, the first AI software engineer. Devin is the new state-of-the-art on the SWE-Bench coding benchmark, has successfully passed practical engineering interviews from leading AI companies, and has even completed real jobs on Upwork. Devin is

5K

10K

45K

2

0

20

@Calclavia

Henry Mao

4 months

@ericneyman Now that you've posted this tweet, it'll be in the training set of the next GPT and get memorized.

1

0

18

@Calclavia

Henry Mao

4 months

How we optimized our servers by 20x 💸 @ 250K monthly active users Several people have been asking why we use Rust, so here's the full story: Back in July 2023 📅, we were serving 80K monthly active users. About 4 requests per second were hitting our servers at any given time

@afonsolfm

Afonso Matos 🇵🇹

4 months

@Davidjpark96 Curious what you guys use rust for?

2

0

2

3

5

17

@Calclavia

Henry Mao

5 months

Moved into our new place in Singapore 🇸🇬! @TheLiyenTan The best thing about SF is the tight-knit startup community that lives in close proximity. We're transplanting the energy from those Victorian-style houses into this renovated heritage shophouse with 16 empty studios.

4

2

16

@Calclavia

Henry Mao

9 years

Officially graduated. Got to throw those fancy hats. :)

5

2

16

@Calclavia

Henry Mao

9 years

This is a revolution. EDX now runs on both MC 1.7 and 1.8. Same mod file. Thanks to @asiekierka for inspiring us to make NOVA. :)

0

2

15

@Calclavia

Henry Mao

5 months

Visas can be stressful, especially when I hear stories from friends trying to work in the USA or get an O1. Fortunately, my experience getting a Singapore visa as an American has been relatively frictionless. Applications are all online, streamlined, and there's no dealing

Tweet media one

2

1

17

@Calclavia

Henry Mao

5 months

Had a great time in San Fransisco! Hosting my last founder dinner this Friday before heading back to Singapore. There'll be co-working, dinner and board games! Details:

4

0

16

@Calclavia

Henry Mao

7 months

Happy to see our academic copilot Jenni help researchers get published and cited!

Tweet media one

0

2

15

@Calclavia

Henry Mao

1 year

@__howardchen When brute force O(n) doesn't work, break the problem down into a hierarchical binary tree O(log(n))!

0

1

13

@Calclavia

Henry Mao

2 years

@then_there_was @sama Evolution is noisy gradient decent. If it worked for us humans, it's plausible for machines.

0

0

13

@Calclavia

Henry Mao

1 year

Went to @sandersted 's OpenAI talk at @weights_biases Fully Connected conference last week. Here's my main takeaway + personal experience working with GPT-4: 👇

Tweet media one

1

3

15

@Calclavia

Henry Mao

4 months

Inspired by the good vibes from the last SF AI Tinkerers I attended (hosted by @mickeyxfriedman ), I'm hosting Singapore's first AI Tinkerers event on August 2nd! This is your chance to showcase your latest AI and LLM technical demos, projects and features to other builders! 👇

4

2

15

@Calclavia

Henry Mao

11 months

Doing some end of year traveling before moving to San Francisco next year!

Tweet media one

2

0

15

@Calclavia

Henry Mao

2 years

@edward_the6 Seems like this will turn into a generative adversial game in the near future...

1

0

13

@Calclavia

Henry Mao

10 years

Graduating soon, then moving to the United States. This is going to be a new adventure!

5

1

14

@Calclavia

Henry Mao

1 year

@SarahChieng I'm very fortunate to be in a similar situation as well with supporting family & friends!

0

0

14

@Calclavia

Henry Mao

1 year

While house cleaning, I found a flyer that @Davidjpark96 and I made way back in 2016. This was for our first startup idea that failed spectacularly. 7 years later, we're still co-founders and our startup @whoisjenniai just reached $100K MRR. It's been a long grind!

Tweet media one

1

0

13

@Calclavia

Henry Mao

9 years

What actually happens when you put lava in a furnace in Minecraft:

1

7

13

@Calclavia

Henry Mao

1 year

@svpino We don't use estimates to exclusively "meet the deadline". Instead, I find it helpful in discovering the true specs of a task. If someone says "8 points" and I say "3 points", we then discuss the gap in feature spec interpretation and come to a consensus.

4

0

12

@Calclavia

Henry Mao

1 year

Writing unit tests can be tedious. Just tried @CodiumAI VSCode Plugin and it generated a full test case for some of my code along with API mocks! Not perfect but it's a pretty impressive start. Worth a try!

Tweet card media

Quality-first AI Coding Platform | Qodo (formerly Codium)

Qodo (formerly CodiumAI) offers quality-first AI tools for developers to write, test, and review code directly within their IDE and Git.

0

2

13

@Calclavia

Henry Mao

10 years

Got blocks to render with the NOVA API! http://t.co/yh3g7XznLM @asiekierka

2

1

12

@Calclavia

Henry Mao

10 months

In 2023, our startup @whoisjenniai grew from $48K to $258K MRR. Last week, I was finally able to take a break and have a relaxing end-of-year + birthday vacation to wrap up an intense 2023. Now back to work for 2024!

Tweet media one

Tweet media two

1

0

13

@Calclavia

Henry Mao

1 year

Took 4 years. We finally hit $2M ARR!

@Davidjpark96

David Park

1 year

How to hit $2M ARR: 1. Struggle for 3 years straight with few results 2. Run out of cash, get funding from Jason Calacanis 3. Fly to Asia to extend runway 4. Thankfully get a little bit of traction 5. Get diagnosed with cancer, kills momentum 6. Get surgery to remove thyroid 7.

Tweet media one

159

67

2K

2

1

13

@Calclavia

Henry Mao

2 years

We grew our startup from $20K to $200K annual revenue (10x) in the last month. 🚀 Here are 3 hacks I've learned as CTO while optimizing our product: 👇

2

6

13

@Calclavia

Henry Mao

1 year

@Thom_Wolf BPE is a compression algorithm at its core. Based on your calculation, this means: GPT4 is ~16.6bit/token Falcon 16bit/token, ~5% more tokens = ~99% GPT4 compression LLama2 15bit/token, ~20% more tokens = ~92% GPT4's compression So GPT4 tokenizer is a superior compressor.

0

0

8

@Calclavia

Henry Mao

11 months

My big accomplishment this week: Successfully explaining the whole OpenAI + Sam Altman drama to my family during Thanksgiving dinner with the limited Chinese that I know. 🤣

2

0

11

@Calclavia

Henry Mao

1 year

@alextreece1 We started as an SEO writing tool. That version of Jenni had no traction or differentiation so we had to pivot to helping academics and researchers.

0

0

9

@Calclavia

Henry Mao

9 years

Finished high school and all exams!! :D

2

0

10

@Calclavia

Henry Mao

9 years

Force field projector chamber. It's multiblock now. http://t.co/3PdmJRRpjD

Tweet media one

5

2

11

@Calclavia

Henry Mao

9 years

The condensed explosive is... back! @LearnToMod #NOVA http://t.co/RZR4f0vNqI

Tweet media one

2

0

11

@Calclavia

Henry Mao

9 years

Force field render is now working... both in MC1.8 and MC1.7. http://t.co/Rw1XNv3yAK

Tweet media one

1

2

11

@Calclavia

Henry Mao

1 year

@Yampeleg Not sure how you define "far behind", but I haven't tried any open-source models that are close to GPT-4's performance on a broad range of tasks. Maybe they're just lacking a quality RLHF/fine-tuning dataset.

1

0

11

@Calclavia

Henry Mao

1 year

Another milestone achieved! We just reached 10,000 paid subscribers on @whoisjenniai !

Tweet media one

1

1

11

@Calclavia

Henry Mao

1 year

When you've worked with your co-founder @Davidjpark96 for 7+ years and now you literally send the same message after a call...

Tweet media one

1

0

11

@Calclavia

Henry Mao

8 months

@ptoleeduh @karpathy Compilers replaced assembly programmers. Current AI will replace today's equivalent of assembly engineers - perhaps front end UI engineers, but not all engineers. Engineers will have to move up to a further level of abstraction. Imo this makes everyone more productive.

1

0

10

@Calclavia

Henry Mao

3 months

#SanFrancisco Founders! I was in SF 🌉 for a few months this year and had a blast meeting many founders. The energy in SF is infectious and I want to bring it wherever I go! If you're an SF founder (or know an SF founder) coming to Singapore 🇸🇬for 3+ months, we've got a place

2

1

11

@Calclavia

Henry Mao

3 months

We're just one week away from Singapore's first AI Tinkerers meetup! We've got community demos discussing the latest insights from testing Llama 3.1 and attempts at tackling the @arcprize . We're calling for more demos! Registration is now at 50% capacity. @jheitzeb

@Calclavia

Henry Mao

4 months

Inspired by the good vibes from the last SF AI Tinkerers I attended (hosted by @mickeyxfriedman ), I'm hosting Singapore's first AI Tinkerers event on August 2nd! This is your chance to showcase your latest AI and LLM technical demos, projects and features to other builders! 👇

4

2

15

2

0

10

@Calclavia

Henry Mao

8 months

Only in SF would I go to an event hosted at a bar to read AI research papers with other academics.

Tweet media one

2

1

11

@Calclavia

Henry Mao

9 years

@CommitStrip if PHP guy says it's badly designed, then it must be really bad... :D

0

0

10

@Calclavia

Henry Mao

1 year

Flying from San Francisco back to Kuala Lumpur to focus on building with my team. Best part of my SF trip was meeting people in the SF AI community - thanks to @GenAICollective , @cerebral_valley , @jowyang for organizing these events! The hype for generative AI is exploding!

Tweet media one

2

0

10

@Calclavia

Henry Mao

6 months

Productive and fun founder coworking session today! @TheLiyenTan @henloitsjoyce @Marcel7an

Tweet media one

1

0

10

@Calclavia

Henry Mao

1 year

Demo-ed Jenni @whoisjenniai at @cerebral_valley along with our new PDF chat feature!

Tweet media one

0

3

10

@Calclavia

Henry Mao

1 year

@jerryjliu0 All of these steps need to also account for latency...

4

0

10

@Calclavia

Henry Mao

5 months

Before flying off to Singapore

Tweet media one

2

0

10

@Calclavia

Henry Mao

1 year

Went to a haunted house ( @hauntumy ) last week for team building at @whoisjenniai ! As the saying goes... nothing builds bonding more than shared fear!

Tweet media one

1

2

9

@Calclavia

Henry Mao

1 year

@Yampeleg Yes, but our brains probably don't predict "tokens" in the discrete sense. It's likely we predict many things in parallel subconsciously via System 1 before it's surfaced to the conscious System 2 brain.

1

0

8

@Calclavia

Henry Mao

5 months

I'm starting a co-living hacker house (~3-4 rooms) in Singapore starting in early/mid-June. Minimum 3-month lease. Looking for startup founders! DM me if you're interested.

2

3

10

@Calclavia

Henry Mao

9 years

Three more days until finishing exams and graduation! Then after that, dedication to programming, indie games, mods... :)

3

0

9

@Calclavia

Henry Mao

1 year

Many GPT users already know to ask it to "think step by step" to get better results. But just pausing and reflecting on this method allows you to truly appreciate its genius. 🧐 LLMs have fixed compute per time step. Each Transformer layer has N layers allocated per token.

0

2

9

@Calclavia

Henry Mao

5 months

@khoomeik Some old approaches trained models one layer at a time, freezing each layer after it's done, before moving on to the next. It works, but it's not as good as backprop. More modern example: Efficient Training of BERT by Progressively Stacking

Efficient Training of BERT by Progressively Stacking

Unsupervised pre-training is popularly used in natural language processing. By designing proper unsupervised prediction tasks, a deep neural network can be t...

proceedings.mlr.press

1

1

9

@Calclavia

Henry Mao

10 years

Who's in favor of electric lights and torches that burn out? :)

7

1

9

@Calclavia

Henry Mao

1 year

Grabbing my last roti and curry puff before leaving Malaysia for some time after building my startup @whoisjenniai here for the last 3 years! Went through the toughest times: - Grew from ~$0 to $150K MRR - Pivoted from an SEO to an academic product - Hit PMF It was a wild ride

Tweet media one

Tweet media two

1

0

9

@Calclavia

Henry Mao

7 months

@swyx They need to stop calling version X.5 Just call it `GPT-2024-04-09` or some clean version number!

2

0

9

@Calclavia

Henry Mao

10 years

Last day of high school. Time flies fast. I've been modding for 3 years!

2

0

8

@Calclavia

Henry Mao

1 year

We just released a new sleeker UI for @whoisjenniai !

Tweet media one

1

2

9

@Calclavia

Henry Mao

2 years

Setting all the research from #EMNLP2022 is giving me a lot of ideas for future research topics in NLG!

Tweet media one

0

0

9

@Calclavia

Henry Mao

1 year

At @weights_biases fully connected conference!

Tweet media one

0

1

8

@Calclavia

Henry Mao

5 months

@y0b1byte L1 is closer to L0 norm. And L0 norm directly optimizes sparsity because it's the count of non zero elements. Only problem is L0 isn't differentiable so it's hard to optimize.

0

0

8

@Calclavia

Henry Mao

10 years

Happy birthday to me :D

5

1

7

@Calclavia

Henry Mao

7 months

@khoomeik Can you elaborate? Intuitively, it feels like if your data is less compressible / harder to compress, you would need more parameters / model capacity to absorb the info entropy within the dataset. Why is the opposite true?

5

0

10

@Calclavia

Henry Mao

10 years

Whoo! Won @ModJamTeam 2014! Thanks for the votes and @electrodynamix for the awesome artwork!

0

0

9

@Calclavia

Henry Mao

1 year

At the @MIT AI Conference in Mountain View! Many top speakers today. #MITAI

Tweet media one

1

0

9

@Calclavia

Henry Mao

10 years

Some new amazing models from @electrodynamix for MFFS! http://t.co/C9XsBJhsTf

4

1

7

@Calclavia

Henry Mao

9 years

This year, my new year's resolution is 3840 x 2160. Happy 2016!

2

2

8

@Calclavia

Henry Mao

9 years

New logo from @theCyanideX calls for a new intro splash screen! http://t.co/Pzmntj0ttv

Tweet media one

1

0

7

@Calclavia

Henry Mao

1 year

@arankomatsuzaki From my qualitative testing, Anthropic's Claude models seem weaker than GPT-4. This could be either due to the RLHF/RLAIF method or pre-training. It's unclear if these approaches actually generalize better in the wild, even if they perform better on academic benchmarks.

0

0

5

@Calclavia

Henry Mao

2 years

Thought-provoking @MLStreetTalk podcast about one of the most mysterious questions in AI: What is consciousness and can machines have it?

Tweet card media

CONSCIOUSNESS IN THE CHINESE ROOM

This video is demonetised on music copyright so we would appreciate support on our Patreon! https://www.patreon.com/mlst Panel: Dr. Tim Scarfe, Dr. Keith Dug...

www.youtube.com

1

1

8

@Calclavia

Henry Mao

1 year

@0xnirlin Depends on what you're looking for. If you're looking for somewhere to focus and not get distracted by other founders/events + a low cost of living with a high quality of life, then it's a great idea.

1

0

7

@Calclavia

Henry Mao

9 months

Back in Malaysia with some tasty banana leaf for $4.

Tweet media one

4

0

8

@Calclavia

Henry Mao

1 year

@dannypostmaa GPT wrappers probably benefit from more competition among LLM providers.

0

1

7

@Calclavia

Henry Mao

10 years

Despite the recent drama on NOVA, I'm still working hard on the project because I really need something to make modding easier.

0

1

8

@Calclavia

Henry Mao

1 year

It's hard to build human connections in remote work. That's why we organize annual team retreats at @whoisjenniai ! Flew the team from UK, Singapore, and India to Malaysia. Meeting some people I've been working with in person for the first time is surreal and exciting!

Tweet media one

2

2

7

@Calclavia

Henry Mao

1 year

I traveled to 6 countries nomadically in the last year. But one thing that I missed was having a strong tech community. The startup journey is hard, and sometimes it's a bit easier when others are in the same boat! Having founders welcome you to their home with food is

Tweet media one

1

0

8

@Calclavia

Henry Mao

9 years

Don't want to touch 1.8 rendering madness and JSON files? Try NOVA's rendering system! http://t.co/PD0fIQA444

1

3

5

@Calclavia

Henry Mao

11 months

Cookies are almost all out! #EMNLP2023

Tweet media one

0

1

8

@Calclavia

Henry Mao

9 months

Building an LLM startup is hard, so we made the ultimate pivot. Now @whoisjenniai is branching out as a bakery! We want to have our cake and eat it too. Here's how we grew from $0 to $3M ARR selling baked goods and cookies. 👇

Tweet media one

1

0

8

@Calclavia

Henry Mao

11 months

Just enjoying the popcorn this weekend

0

0

7