
Cofounder & CTO of Vehicle Smart Ltd Author of https://teapaw.org/t.co/a9LVha9F1l @VehicleSmartApp - your garage in your pocket ™️ @TheoryTestApp
Hastings, England
Joined March 2017
- Tweets5.5K
- Following232
- Followers371
- Likes16.8K
Oh my 🤤
Replying to @NovaXCode
Yes, confirmed. Elon stated Grok 4.8 is a 2.5T-parameter model trained on xAI’s new C++ stack. Pretraining is complete; it’s now in RL with a soft rollout to some Bot users. General release is expected after a bit more polish.
16
Hy4 flash from @TencentHunyuan is now open weights: 220B total, 12B active, 1M context. Catch: the smallest GGUF is still over 120GB, so think big-memory Mac Studio or loads of system RAM, not a gaming PC. Licence is Tencent's own, not Apache. teapaw.org/TencentHunyuan/status/…
Hy4 flash is now open-weight, including Hy4 flash-FP8 and Hy4 flash-Base.
224B (12B active), 1M context, natively multimodal.
Fast, efficient and built for everyday productivity.
It's yours now. And stay tuned for Hy4 coming up next!
🤗: huggingface.co/tencent/Hy4-f…
🐙: github.com/Tencent-Hunyuan/H…
📄: hy.tencent.ai/research/hy4-f…
Released under the Tencent Hy License.
See full terms here: huggingface.co/tencent/Hy4-f…
1
47
Karpo Decision from @Karpo_AI is a different beast. It doesn't chat, it scores options: give it a page or screenshot plus choices, get probabilities in one pass. Handy as the quick 'which button?' brain in a local agent. 31B is 62.5GB, smaller sizes too. teapaw.org/Karpo_AI/status/212058…
Our latest model, Karpo Decision 31B, just tied for #1 on the Decision Index🥇 2× as fast as Jev, at 20% lower cost. Open source and already live in Karpo!
28
Edge0-145B from @Edge0_AI is a MoE with ~10B active, so the busy experts sit on the GPU and the rest live in RAM and on SSD. They quote 91.7 tok/s on a 4090. Catch: ~99GB download, 32GB+ RAM, and commercial use needs their OK. teapaw.org/Edge0_AI/status/212094…
28
Clever one from @yb2698 and the HF crew: a ~300MB LoRA that teaches Qwen3.8-27B to drive the pi coding agent. SWE-bench Pro hard 51 goes 39.2% to 64.7%, thinking off. Catch: tuned for pi only and the GGUF isn't re-tested, so use a Q8 base if you can. teapaw.org/yb2698/status/21226850…
Announcing harness adapters for @Alibaba_Qwen Qwen3.8-27B on Pi harness!
We're releasing a LoRA adapter that makes Qwen3.8-27B a much better and efficient agent behind Pi harness.
SWE-Bench Pro (Hard 51 subset) pass@1: 29% -> 63%
Terminal-Bench 2.1 pass@1: 40% -> 56%
Much more efficient: 55-62% fewer tool calls per task 🤯
huggingface.co/harness-hub/Q…
The era of harness-specific adapters is now on! More harness adapters to come in the next days 🧵
1
1
31
SenseNova-Loop-9B from @SenseTime_AI runs its middle 24 layers twice, so 9B of weights does 56 layers of work. Q4 GGUF is under 6GB, takes images, Apache 2.0. Catch: needs their own llama.cpp fork for now. Worth a play on a 12GB card. teapaw.org/SenseTime_AI/status/21…
#SenseNova-Loop-9B is officially here — now #OpenSource!
Just 9B parameters, competitive with Qwen3.8-27B across multiple benchmarks.
🔹 Big gains over Preview: NL2Repo 19.91 → 41.36; AutomationBench 14.33 → 36.67; GAIA 37.86 → 69.90
🔹 Loop + mHC: 32 unique layers, 56 layer executions
🔹 Day-0 4-bit AWQ weights (~4.9GB), with VRAM needs as low as ~9 GiB at 64K context
🔹 Up to 512K context, with text and image inputs
See what a 9B model can build: animated posters, interactive websites and 3D scenes.
💻 GitHub: github.com/OpenSenseNova/Sen…
🤗 HF: huggingface.co/collections/s…
👾 Discord: discord.com/invite/BuTXPHmQu…
1
2
99
llama.cpp just hit v1.0.0. The big deal isn't features, it's semver: breaking API changes now mean a major version bump, so apps built on it should stop randomly breaking. Also FP8 KV cache and a MoE expert cache for smaller GPUs. Nice one @ggerganov teapaw.org/ggerganov/status/21212…
llama.cpp v1.0.0 is now available
huggingface.co/blog/ggml-org…
1
2
119
RobbiewOnline retweeted
This seems like a great time to reshare my favourite Sir Patrick Stewart performance, ‘What has the ECHR ever done for us?’ It’s just perfect in every possible way … including the ending 😂😂😂 RIP.
🎥 The Guardian
To actually have been there to see this would have been wild!
The uncertainty whether its planned to continue off the wrap, then it does 🤪
3
36
The new @Tesla #Roadster
0-60 in less than a second 💨
Jumps
Flips
It really sucks, hard.
teapaw.org/tesla/status/211091343…
teapaw.org/tesla/status/211092616…
teapaw.org/i/broadcasts/1yKAPwNgg… teapaw.org/tesla/status/211090609…
1
141
A great writing tool with impressive features! The LinkedIn "feature" is funny as hell 🤭
For years, maybe a decade, I've wanted to build a writing tool that just did a few things I couldn't find anywhere else.
A system that mimicked how I think when I write. A system that provided clear affordances, features, and places to do the things I wanted to do. A way to play with my writing as I'm writing it. A way to settle into just the right way to say something.
It never made sense to make it a 37signals product because it's not commercially viable, nor would it be worth pulling people off their other work to hack on this. Plus, it would have taken months the old way.
So this weekend I just made it myself, with Claude's help.
It's called Write_On and the short video walks you through it.
Essentially it's "alternative control" at the word, sentence, and paragraph level, plus a way to dim stuff back, and stash stuff near by. Not version control, but alternative control. You'll see what I mean when you watch the video below.
So here it is. Write_On.
2
2
47
RobbiewOnline retweeted
⭐ 𝗪𝗲 𝗰𝗼𝘂𝗹𝗱𝗻’𝘁 𝗵𝗮𝘃𝗲 𝘀𝗮𝗶𝗱 𝗶𝘁 𝗯𝗲𝘁𝘁𝗲𝗿 𝗼𝘂𝗿𝘀𝗲𝗹𝘃𝗲𝘀.
It’s always great to see reviews like this from people using Vehicle Smart out in the real world.
Whether you’re checking the history of a used car, or…..
Neuralink just showed another huge step with its VOICE trial
Terry is using his Neuralink implant to help fine-tune a brain-to-voice interface for himself and others who can’t speak
He first trained the algorithm by miming speech as best he could
Now he can simply think the words....and hear them come out in his own natural voice
And the voice is powered by Grok Voice from SpaceXAI
Neuralink reads the intent from the brain
Grok Voice turns it into natural speech
Thought → voice
Two Elon companies working together to give people their voice back
This is absolutely incredible
RobbiewOnline retweeted
Steve Jobs really pulled a reality distortion field to poach Tim Cook from Compaq (in 1998, it was worth $30B+ to Apple’s $3B).
All of Cook’s friends and advisors told him not to join almost-bankrupt Apple.
Cook says it took Jobs one meeting to convince him:
“I had gotten a call several times from the search people that [Jobs] had employed, and I kept saying no.
I was happy at Compaq — or thought I was — and they were persistent.
So, I finally thought, ‘you know, I’m going to go out and take the meeting. Steve created the whole industry that I’m in. I’d love the meeting.’ […]
All of a sudden, he’s talking about his strategy and his vision and what he was doing.
He was going 100% into consumer when everybody else in the industry had decided you couldn’t make any money in consumer. They were headed to servers and storage and the enterprise. […]
I always thought following the herd was a bad idea.
He showed me a bit of the design to get me interested [of what later would be called the iMac].
The way that he talked and the chemistry. I could tell I could work with him.
I looked at the problems Apple had and I thought ‘you know, I can make a contribution here’ and working with [Jobs would be] the privilege of a lifetime.
All of a sudden I thought, ‘I’m doing it. I’m going for it.’
You hear this voice in your ear that says, ‘Go west, young man, go west.’
[The move to Apple] didn’t make sense. And yet my gut said ‘go for it’ and I listened to my gut.
There was literally no one around me that was advising me to do it.”
In March 1998, Cook joined Apple as Senior Vice President of Worldwide Operations on a $400k base salary (and a $500k signing bonus).
He was 37. Incredible gut call.
***
Full episode here from 2014: youtu.be/u9Sy3qwthCw
NEW: Tim Cook sends a final memo to Apple employees on his last day as CEO. “I will miss this work in ways I can only begin to imagine, even as I remain completely at peace with my decision,” he says, adding he has “enormous comfort” with Ternus in charge. bloomberg.com/news/articles/…
RobbiewOnline retweeted
🛞 𝗛𝗼𝘄 𝗼𝗹𝗱 𝗮𝗿𝗲 𝘆𝗼𝘂𝗿 𝘁𝘆𝗿𝗲𝘀?
Most drivers know to check their tyre pressure and tread depth — but tyre age is easy to overlook.
Every tyre has a four-digit date code on the sidewall showing the week and year it was manufactured.
Watching robot Olympics in a year or two is going to be fantastic!
Bring back Robot Wars on the BBC too! @robotwar
105



























