
General Partner @TRIVEDIGI | Advisor & Mentor @LoyalVC @TzApac @OVioHQ | Founder, Investor, Advisor | Fintech, Web3, AI | All views expressed are personal
Joined February 2023
- Tweets89
- Following141
- Followers68
- Likes178
ShawnT retweeted
Riel invited you to Call Me Sensei @callmesenseieng 💜! Use my referral code 1CGZQYBL to get bonus rewards when you sign up. 🎁
callmesensei.app/r/1CGZQYBL
ShawnT retweeted
Getting there!
After about 10 hours, it's quite clear that Opus 5.5 will NEVER get to a perfect 3D model without significant intervention, at least with the first character and without skills.
Look at the eyes for example, I had to spend 2+ hours so far to get it to become closer to the reference. Somehow Opus 5.5 are blind to these minor details or maybe it just doesn't care enough.
The second finding is that you simply CANNOT use subagents. If you want to use subagents, they need to be scoped to literally focus on a single body part or it will go into an INFINITE loop of slop and you have to throw everything away since it just never move any closer.
I believe it will still require me babysitting it for another 2 hours to get to where I need the eyes to be since it kept claiming that it's done, but it's just not done. So far, I think it's been 12 cycles.
Getting there, but not quite, yet. The head is really the most difficult part so far.
ShawnT retweeted
This is the outcome so far after 3 iterations. This game wass supposed to be a 2D side scrolling game, but using sprites were painful and using Spine2D has proven to be too much for current generation of LLMs.
So I went 3D with fixed 2D perspective camera instead and now on @threejs.
A quick summary of what Opus 5.5 is now able to do quite effectively:
- Rig character
- Program motion/animation
- Analyze frame-by-frame and correct the animation
- SFX and sound alignment
- VFX
I would say Opus 5.5 is heavily optimized for game development. I have used it 80% for general coding and 20% for game development (5 completely different games) and you can really feel the difference.
I also had a chance to test ToonLab quite extensively and so far Opus 5.5 also gave it a praise randomly by itself, but there are definitely things to improve.
Nana's Kendo Legends will be available in @callmesenseieng probably in a week or so. I will expand it to cover 10 levels for the first release.
The most interesting part about Opus 5.5 is that it feels as if it's came with its own set of tools. Here's one example. The first round of implementation was shit since the enemy were very stiff. Sure enough, 5 seconds later, it brought up this playground and tested every motion. Basically, it knew exactly what to use as if it was packaged with Opus 5.5.
ShawnT retweeted
me: i literally cannot do this anymore
rem: *texts*
me: okay i can do it for like 5 more minutes
#CallMeSensei #AnimeGame #OtomeGame #DatingSim #VisualNovel
ShawnT retweeted
what if learning a new language felt like stepping into an anime?
meet the Senseis you can talk to, call, go on simulated dates with, and play games with—all while practicing a new language.
Call Me Sensei is now in Open Beta 💜
callmesensei.app/quiz
ShawnT retweeted
Since everyone have been debating about how fast GPT-6 Astra has been draining their account and I just saw that there is a way to extract this info. Here's my exact usage since Astra became available. The estimate is very similar to the 30 hours of usage time per 100% weekly limit I shared earlier as a reply to another discussion.
Quick point to this is that you can still use GPT-6 Astra High effort almost 100% of the time and should still last like 4 days unless you run things in parallel.
--
Codex usage across roughly 29 hours on September 5–6:
- 1.145 billion recorded tokens
- 6,298 model requests
- 5,515 logged outer tool calls
- 95.84% of input tokens cached
Astra handled 4,000 requests and 876.47M tokens. About 91% were High effort. Only 102 Astra requests ran at regular speed; the other 3,898 were Fast.
The estimated API retail value was $3,481.85, excluding usage we couldn’t price.
I consumed approximately 217% of a weekly allowance across accounts and resets. Valuing each full allowance at $50 gives $108.50 of allowance value. About 32× less than the API equivalent.
The time comparison is what stood out:
One model request at a time, at recorded speeds: 25.6 hours.
The same workload at regular speed, assuming Fast delivered a 2.5× speedup: approximately 61.8 hours.
Per full weekly allowance consumed:
61.8 ÷ 2.17 = 28.5 hours of regular-speed-equivalent model time per 100% limit.
That’s equivalent processing time, not a guaranteed regular-mode quota or human work time. Tool overhead is extra.
These are estimates from local logs and stated assumptions, not an official bill. The billion-token total includes repeated cached context. Not a billion newly generated tokens.
Update: this went from 16 to 35 different people reporting unusually fast Astra usage burn in about an hour.
I also had Codex review my local usage logs, then had ChatGPT Pro running Astra do a deeper analysis and generate the infographic below.
Over the last 48 hours, my usage shows:
• 2.11× more model calls per day
• 83.9% more tokens per day
• Astra was 64.8% of recent tokens but 87.9% of the estimated standard-rate cost
• Cache hit rate stayed basically unchanged at ~96.7%
So this does not look like a simple caching failure.
I'm still not saying I've proven there's a bug. But when my own usage shows a major shift and 35 independent users are reporting unusually fast burn, I think this has moved well past "Astra is just more expensive."
However it might need an actual look @thsottiaux
ShawnT retweeted
Call Me Sensei is officially in OPEN BETA! 💜 Call Me Sensei is an immersive language learning dating sim where you can build relationships with your Senseis while practicing language through natural conversations and experiences that become part of your daily life.💬 #OpenBeta
ShawnT retweeted
Just published a song for the first time. The song is for @callmesenseieng and it's called「声の先へ」ー Koe no Saki e ー Beyond the Voice.
The process was way easier and cheaper than I thought using @DistroKid. If all goes well, the song will be available on all major platforms sometime within the next 72 hours.
ShawnT retweeted
From chatbots answering questions to AI agents taking action. What actually changes for everyday users?
At the ACTIV8 Retail Stage, Su Fen Goh (@_sugoh) from Chad, Jake Ong (@Jake_Ong9) from PennyWhale (@pennywhaleio), and Shawn Tan (@gimmemoat) from Call Me Sensei (@callmesenseieng) joined moderator Hoa Ha Phuong (@Hoa_lp_Ha) from Spores Network (@Spores_Network) for the panel:
“From Chatbots to AI Agents: What Changes for Everyday Users?”
A practical discussion on how AI agents could reshape daily digital experiences, from smarter automation to more personalised user journeys.
ShawnT retweeted
14:15–14:45 | From Chatbots to AI Agents: What Changes for Everyday Users?
How will AI agents move from hype to actual daily utility?
Jake Ong (@Jake_Ong9), Su Fen Goh (@_sugoh) and Shawn Tan (@gimmemoat) join the discussion, moderated by Hoa Ha Phuong (@Hoa_lp_Ha).
ShawnT retweeted
🎤 Speaker at #WikiEXPOHK2026
@gimmemoat is General Partner at @TRIVEDIGI , focusing on investments in Web3 and AI. He is also an advisor to TZ APAC and Loyal VC, a mentor with Outlier Ventures, and founder of Hyperbond Studio, creator of the AI language-learning platform Call Me Sensei. Shawn holds an MBA from INSEAD and regularly teaches AI and digital banking as an adjunct lecturer.
🌐 Join us: lu.ma/WikiEXPO
#WikiGlobal #WikiFX #WikiBit #WikiGold #TRIVE
ShawnT retweeted
GPT-5.6 Sol confirmed to be an extremely good model
Beginning July 20, Claude Fable 5 will be included in all Max and Team Premium plans, at 50% of limits.
Pro and Team Standard users will continue to have access to Fable via usage credits, and will receive a one-time $100 credit.
Demand for Fable has been challenging to predict, which is why we rolled it out to subscription plans in stages, extending access several times as we secured additional capacity.
ShawnT retweeted
Text-to-animation definitely still has a long way to go, but you can now iteratively prompt with a model like @AnthropicAI Fable 5 to get the animation you want. The experiment was only for the freestyle swimming part.
Summary:
- Task: Freestyle swimming with head turning to get some air every 4th stroke (ended up with every 2nd stroke, which is fine also)
- 3% of weekly Fable allowance was used
- Estimated cost: $1.20
- 7 iterations because Fable's vision is so terrible (please fix it @AnthropicAI). I had to keep taking screenshots myself and then describe every painstaking details of what's wrong with it, but it did get there eventually.
And yes, all the shaders, water mechanics, tree, grass, etc. you see here will be open sourced soon (1-2 weeks). This is being developed as a mini-games engine based on @threejs for @callmesenseieng (Open Beta available soon).
ShawnT retweeted
We built language companions to help people learn languages.
Now our own team won't stop talking to them. 💀
Download Call Me Sensei and meet your Sensei today. 💜
#CallMeSensei #LanguageLearning #StartupTok #Anime #DatingSim
ShawnT retweeted
Proud to have played a major role in making this happen. Something is brewing 🟦 – at Kuala Lumpur, Wilayah Persekutuan Kuala Lumpur
This quoted post is unavailable.
ShawnT retweeted
The level of skills and creativity in our @callmesenseieng team is getting to a very interesting point. We have been working hard to upskill ourselves and also created numerous tools to help save time and ensure quality.
This started off from me giving a training on how to use AI tools at the end of January. Now I feel like I'm going to be the one learning instead.
Follow along for some crazier content.
Do you shread🏂? Ren does…talk to him in the Call Me Sensei App📲✨
#callmesensei #renkanzaki #snowboarding #スノボ #animeasethetic
ShawnT retweeted
🎉 TEAMZ Summit 2026 Speaker Announcement
🎤 Shawn Tan @gimmemoat
🔹 Co-founder and CEO
🔹Hyperbond Studio
💼 Shawn is a serial entrepreneur turned investor and advisor with over a decade of experience in venture investments, strategy, and zero-to-one management. As Co-founder and CEO at Hyperbond Studio, Shawn launched Call Me Sensei, an AI EdTech platform revolutionizing language learning, built upon the insight that the best way to learn a language is by dating someone who speaks it.
📅 April 7–8, 2026
📍 Happo-en, Tokyo
🎟️ Tickets: tickets.teamz.co.jp/
🌐 teamz.co.jp
#crypto #Web3 #AI #TEAMZ #Blockchain #NFT #DeFi #Investor #Web3 #Summit #Event
ShawnT retweeted
This is basically Call Me Sensei, but in our case, it’s done in the Anime themed world of CMS.
Introducing Simile.
Simulating human behavior is one of the most consequential and technically difficult problems of our time.
We raised $100M from Index, Hanabi, A* BCV, @karpathy @drfeifei @adamdangelo @rauchg @scottbelsky among others.
ShawnT retweeted
Text-to-Speech at the level of quality we were trying to achieve was beyond reach from cost and performance perspective before we started working with @MiniMax_AI. It’s been a pleasure working closely with their team to continue to improve so that our senseis feel as human as possible.
As of today, Minimax TTS 2.8 was enabled along side our Sensei Engine V4. Give it a try if you have no already. Open alpha version of the app is available on both iOS and Android.
This Post is from a suspended account. Learn more
























