Can I please get a model that's the lovechild of Opus 5.5 and Fable 5.2 in terms of ability/behavior?
1
21
There's something extra satisfying about cracking a hard problem entirely with LLMs running on your own hardware.
Today I decrypted a locked-down device, mapped it end to end, and found a full remote path for my own code.
GLM 5.3 Flash is seriously amazing.
1
31
This is so much faster than what I was used to that I had to double-check I hadn’t accidentally selected a cloud GLM 5.3 Flash in OpenCode.
Awesome work @MiaAI_lab @plotarmordev @ashxhart!
Run GLM 5.3 Flash EXL3 with TensorFold ⚡️
This is a completely new recipe that ushers a whole new level of performance for @NVIDIAAI 2x DGX Sparks.
Conservative default for stability:
- 1M context by default
- 2.7M KV cache pool (!)
- Yes, it's not a typo - 2.7M KV in just two Sparks
- 4 concurrent streams by default
Performance:
- 60 tok/s on prose, single stream.
- 108 tok/s on prose, 4 concurrent streams.
~1,950 prefill tok/s for most context lengths.
Stress-tested to handle a variety of workflows.
This is by far the BEST model to run if you have two DGX Sparks. Extremely smooth experience!
Expect further improvements!
Thanks to @ashxhart for developing such a powerful engine! TensorFold will be used in many of my upcoming recipes.
Get it here:
github.com/MiaAI-Lab/GLM-5.3…
1
101
Wild that the 256GB M5 Ultra deliveries are out to the end of January already. Everyone is desperate for compute.
1
1
105
Frontier AI models seem to be getting smarter while simultaneously becoming worse communicators.
If I have to mentally decompile your explanation before I can understand it, you’ve failed at the explanation.
1
1
30
Matt Kelch retweeted
To summarize: HuggingFace got autonomously compromised by a model from an American company. HF then tried to use American frontier model(s) to defend themselves, but were blocked by guardrails. HF then had to turn to open source Chinese models to defend themselves from another American company.
We're partnering with @huggingface to investigate an unprecedented security incident.
Cyber-capable OpenAI models compromised Hugging Face production during a benchmark evaluation.
Sharing preliminary findings to help defenders understand emerging risks:
openai.com/index/hugging-fac…
Matt Kelch retweeted
I have also run this fun thought experiment! More of the world than many might imagine could run on outdated hardware if software optimization was truly a priority, and market price signals on scarce compute would make it happen. Rebuild all the interpreted microservice based products into monolithic native codebases! Innovative new products would get much rarer without super cheap and scalable compute, of course.
I continue to be amazed at just how consistently recognizable these are. Here’s one I haven’t seen posted yet.
53
Replying to @TouchTunes
@TouchTunes thank you for finally fixing the most annoying quirk about your app. Loving the fact that playing a song no longer immediately refreshes the “Your recent plays” list — it’s been driving me nuts for years.
1
53
Matt Kelch retweeted
My new favorite thing - Bing's new ChatGPT bot argues with a user, gaslights them about the current year being 2022, says their phone might have a virus, and says "You have not been a good user"
Why? Because the person asked where Avatar 2 is showing nearby
Anyone else wondering if the @Spotify downtime tonight somehow lead to @TouchTunes downtime? 🤔
1
215
Matt Kelch retweeted
Replying to @elonmusk
I can't post it yet - just waiting for last legal clearances from our lawyers. It's insane that, as a journalist, I even need to worry about this, but that's the climate here.
But I'll detail the contents on our show at 7pm, and yes, many of it is directed at Twitter.
Matt Kelch retweeted
Hmmm. We move our satellites on average once a week and don't put out a press release to say who we maneuvered around... cnet.com/google-amp/news/spa…































