Lewis N Watson

8.7K posts

Lewis N Watson banner
Lewis N Watson

Lewis N Watson

@LewisNWatson

PhD Student in Conversational Visual Dialogue. VP ENUSEC. CTF Dev. Citizen/Env Science. Somehow Sentient Meat. Scottish. Disclaimers in extended bio.

ルイス・ワトソン Beigetreten Nisan 2020
1.1K Folgt645 Follower
Lewis N Watson
Lewis N Watson@LewisNWatson·
@sudoingX preprint is from april 2025 surely they’ve tested more since
English
0
0
0
76
Sudo su
Sudo su@sudoingX·
people keep asking me about turboquant. google's KV cache compression. 3-bit, zero accuracy loss on their benchmarks. here's what nobody's pointing out. google only tested this on 8B models. nobody knows what happens above that. does 3-bit KV cache hold at 27B, 35B, 70B or does it break at scale? i need these answers nobody's has yet. i already have baseline KV cache data across multiple models and architectures on the same hardware. the before data exists. the experiment list is long. when results come in i'll publish everything raw, no filters for you all, i am in deep.
Anes Hujević@aneshujevic

@sudoingX @sudoingX can’t wait for turboquant to be implemented in these, will you be testing them also?

English
33
12
449
26.7K
Lewis N Watson retweetet
James Reed
James Reed@jamesr66a·
James Reed tweet media
ZXX
4
75
941
17.2K
gabriel
gabriel@gabriel1·
hello friends
English
575
3
1.2K
95.7K
Lewis N Watson retweetet
Mistral AI
Mistral AI@MistralAI·
🔊Introducing Voxtral TTS: our new frontier open-weight model for natural, expressive, and ultra-fast text-to-speech 🎭Realistic, emotionally expressive speech. 🌍Supports 9 languages and accurately captures diverse dialects. ⚡Very low latency for time-to-first-audio. 🔄Easily adaptable to new voices
English
137
568
4.2K
722.1K
moondream
moondream@moondreamai·
VLMs too slow for production? Not anymore: 46ms end-to-end inference, 60+ fps on a single H100. Introducing Photon, Moondream's inference engine. Runs on everything from edge to server. moondream.ai/blog/photon-re…
English
36
98
1.3K
212.4K
Lewis N Watson
Lewis N Watson@LewisNWatson·
@stevibe 9b -> 27b for the footprint difference is a pretty decent balance
English
0
0
0
111
Lewis N Watson retweetet
stevibe
stevibe@stevibe·
Which local models can actually handle tool calling? I built a framework to find out. 15 scenarios. 12 tools. Mocked responses. Temperature 0. No cherry-picking. Tested every Qwen3.5 size from 0.8B to 397B, and since some of you asked after the distillation tests: yes, I included Jackrong's Qwen3.5-27B-Claude-4.6-Opus-Reasoning-Distilled too. Only two models went all green: the 27B dense and the distilled 27B. The 397B? Failed two tests. The 122B? Failed one. The 35B? Failed two. The timed-out results — mostly on the smaller models, are cases where the model got stuck in a loop, repeating the same tool call until it hit the 30-second limit. The test that exposed the most models: "Search for Iceland's population, then calculate 2% of it." Simple, but 35B, 122B, and 397B all used a rounded number from memory instead of the actual search result. They didn't trust their own tool output. Small models hallucinate data. Big models ignore data. The 27B just threaded it through.
English
101
223
1.9K
358K
Lewis N Watson retweetet
Hugging Models
Hugging Models@HuggingModels·
Qwen3.5 0.8B running real-time video captioning on a Mac Studio M2 Ultra. <1s per frame. 269 frames from a 3m49s video. Streaming descriptions as it plays. Pause anywhere, it actually understands the scene. ~1GB model. Local AI is getting unreasonably capable. Video credit: @stevibe
English
54
281
2.9K
249.5K
Lewis N Watson retweetet
Techlore
Techlore@TechloreInc·
🚨 UK: The govt wants to ban VPNs, social media & AI chatbots in various contexts. Your silence is a yes. Make your voice heard before May 26. Fill out the survey and let them know your views: gov.uk/government/con…
Techlore tweet media
English
83
717
1.8K
138.1K
Lewis N Watson retweetet
Obscura: The Privacy-first VPN
I don't even need to say anything, I'm just tired of this.
Obscura: The Privacy-first VPN tweet media
English
104
1.3K
5.2K
155.1K
Björn Fri
Björn Fri@bjornfri2020·
@BigBrotherWatch It’s also huge. The entire operating system is being swapped out. It’s that deeply integrated.
Björn Fri tweet media
English
13
17
129
27.6K
Big Brother Watch
Big Brother Watch@BigBrotherWatch·
🚨NEWS: UK iPhone users must now prove their age or lose full internet access "It is absolutely outrageous that, overnight, Apple has put a chokehold on Britons' freedom to search the internet, access information and use apps unless they provide sensitive ID documents. This means 35 million Brits who have paid hundreds or even thousands of pounds for Apple tech suddenly now have a child's device unless they comply with invasive demands for personal information that go far beyond what UK law requires. Apple has crossed the Rubicon with this software update which is more like ransomware, holding customers hostage to ID demands that are invasive, exclusionary and unnecessary. Children's online safety is vital but requires better parental controls and thoughtful tech responsibility - not sweeping, draconian, shock demands by foreign companies for all of our IDs and credit cards." - Silkie Carlo [@silkiecarlo]
Big Brother Watch tweet media
English
523
1.9K
5.2K
951K
Wildminder
Wildminder@wildmindai·
NVIDIA says: no more "brute force every pixel" of video understanding. AutoGaze- identifies and removes redundant video patches before they enter a Vision Transformer. Now we can processes 4K long-video in real-time. Works with SigLIP2 and NVILA. autogaze.github.io
English
75
163
2.4K
287.3K
OpenAI
OpenAI@OpenAI·
The more AI can do, the more we need to ask what it should and shouldn’t do. OpenAI researcher @w01fe joins host @AndrewMayne to explore the Model Spec, the public framework that defines how models are intended to behave. They break down how it works in practice, from the chain of command that resolves conflicting instructions to the way it evolves over time through real-world use, feedback, and new model capabilities.
English
330
115
1.2K
181.3K