I spent a few hours today poking at the personality neurons @_lyraaaa_ found in Qwen3.5-4B. I wanted to know whether one neuron really carries something like a personality, or whether pushing it hard just knocks the model off balance.
All of it ran on free Kaggle GPUs 🧵
Not everything held up. I guessed the turned-down alien might become a ghost, but only 4 of 40 replies did, short of the bar I'd set. lyra's single sarcasm neuron gave sarcasm in 6 of 32 replies, and her five-neuron mix gave 28. These are keyword counts until a blind judge runs.
It all runs on Kaggle's free GPUs, so anyone can rerun it or try their own neurons. Code, prompts, every raw reply and the scoring script: github.com/shayansanjideh/ne…
Next I want to see whether Claude can find neurons like these on its own, using @treetowntree's Neuron Zoo.
My first impression of the DeepSeek app is uhhh… not too good…
FYI I did different combinations of enabling “Think” and “Search”.
The biggest thing holding me back from trying Instict is the fact that everything is in a single iMessage thread.
I feel like it'd drive me nuts if I didn't have the ability to separate different topics by new chat.
These guys have been cooking for a while now.
We now have Opus/Sol tier intelligence models able to run on hardware the average person has laying around.
Would not be survived if we see Fable/Astra tier models available locally by the end of the year.
Today, we’re announcing Ternary Bonsai 2 27B.
Based on Qwen3.8 27B, Bonsai 2 27B is 9x smaller than its full-precision counterpart while retaining 98.2% of its aggregate benchmark performance.
Two months after the first Bonsai 27B release, the biggest change is quality. The footprint remains 5.9 GB, but the gap to full precision has narrowed materially, with particularly strong gains in agentic coding, multimodal reasoning, and long-horizon tool use.
Ternary Bonsai 2 27B is available today under Apache 2.0.
Shayan retweeted
"muse is the best app - AI or otherwise - I've used since ChatGPT in terms of bringing back that original 'wow' factor."
you're making me emotional 🥹🥹🥹
Muse is the best app - AI or otherwise - I've used since ChatGPT in terms of bringing back that original "wow" factor.
It just... works
The UI/UX is extremely warm and friendly, the AI itself feels very intelligent, and it needs virtually zero handholding.
My biggest gripe with ChatGPT's Work / browser use is the fact that so many actions are blocked, but for whatever reason, Muse's works just fine out the box, no hacky workarounds or anything.
I've been recommending it to my friends working on their local small businesses, and they're blown away by how intuitive it is.
I also think this is truly the future of AI for the vast majority of people: a warm, personal assistant, that just... works.
Also there’s a certain poetic aspect to the messaging around this.
If you go on @alexandr_wang or @MattPRD’s feed the past few days it’s all ways that Muse is automating boring and tedious work, helping people save/earn money, etc. Only ways it’s making people’s lives easier and better.
None of that doomer defeatist bullshit the other labs are preaching.
The Zuck redemption arc is real.
Muse is the best app - AI or otherwise - I've used since ChatGPT in terms of bringing back that original "wow" factor.
It just... works
The UI/UX is extremely warm and friendly, the AI itself feels very intelligent, and it needs virtually zero handholding.
My biggest gripe with ChatGPT's Work / browser use is the fact that so many actions are blocked, but for whatever reason, Muse's works just fine out the box, no hacky workarounds or anything.
I've been recommending it to my friends working on their local small businesses, and they're blown away by how intuitive it is.
I also think this is truly the future of AI for the vast majority of people: a warm, personal assistant, that just... works.
Muse is the best app - AI or otherwise - I've used since ChatGPT in terms of bringing back that original "wow" factor.
It just... works
The UI/UX is extremely warm and friendly, the AI itself feels very intelligent, and it needs virtually zero handholding.
My biggest gripe with ChatGPT's Work / browser use is the fact that so many actions are blocked, but for whatever reason, Muse's works just fine out the box, no hacky workarounds or anything.
I've been recommending it to my friends working on their local small businesses, and they're blown away by how intuitive it is.
I also think this is truly the future of AI for the vast majority of people: a warm, personal assistant, that just... works.
Last month I wrote about how we can build a positive and safe future for everyone: meta.com/thefutureisforevery…
Every lab has the responsibility and incentive to move at the pace required to train its models safely, and the ability to take its own actions to ensure that happens.
The reality is:
- People won't want to use agents that are misaligned with them and that don't do what they ask, so labs have a strong natural incentive to make their models more aligned.
There is a lot of debate about slowing progress on capabilities until alignment catches up. My view is that trust and alignment are quickly becoming the most important capabilities that will differentiate agents and models. Any lab that doesn't focus on alignment will fall behind.
- Labs face significant liability if their models cause harm, so they have a strong incentive to prevent this as well.
Meta delayed shipping Muse for several months to focus on safety and security. We didn't call for everyone else to do this before we would. We just did it as part of our day-to-day work because it was clearly the right thing for people and for us. I'm proud of the security foundations we've built.
- Engaging independent evaluators and advisors is industry best practice. MSL already does this today in several areas because it helps produce better work. Other labs can just do this too. In general, it would be helpful for there to be a larger and more diverse ecosystem of evaluators.
- Committing the significant majority of compute towards serving people rather than racing towards recursive self-improvement is one of the best ways to ensure we develop this technology safely. Meta has made this commitment and other labs can do this as well.
I believe the key to building a positive future for everyone is maintaining the right balance of power. This is within our power to do.
