Nonplussed Netizen retweeted
Hearing this week that personal assistants cost $3k-$7k per user per year, depending on scale effects / vertical integration / model selection. Implications/questions:
- Consumer subsidies will be a huge factor; people don’t care about privacy, they want utility and they want it for free.
- Model costs are deflating, but not fast enough for many startups. The dark horse is that Jev may rapidly deflate computer-use costs 100x.
- An interesting counter-position would be building an expensive and narrowly focused agent — perhaps combining proprietary supply (Carbone tables!) with a unique POV (“no one goes there anymore, go to The Grill instead”).
- Social is wide open but hinges on a key design challenge: how can agents increase social cohesion and not feel invasive in a group chat? A working social agent would rightly attract an avalanche of capital and be hard for incumbents to replicate.
- Finally, there are vertical focus areas like financial services where the LTVs outweigh the subsidy needed and are outside the near-term competitive focus of the incumbents. If it were me, I’d focus here — Credit Karma as an agent >> Credit Karma as an app.
Overall, it’s still very, very early, and we may very well be in the MySpace era of social or iPhone 2010. Hang onto your hats and keep building!!
Nonplussed Netizen retweeted
Replying to @AndrewCurran_
IMHO, any societal structure that can survive only through friction of information flow doesn't deserve to keep existing in the first place.
Replying to @maximusgrave
are we serious
Nonplussed Netizen retweeted
Replying to @realDonaldTrump
Nonplussed Netizen retweeted
This is basically an Opus 4.5/4.6 level model that will work on anything with 8 Gb RAM
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.
Nonplussed Netizen retweeted
We fine-tuned @thinkymachines' Inkling-Small to fix LaTeX compile errors inside the editor, in under a second. On real errors it matches Claude Fable 5.1, six times faster and at about 1/40 of the cost.
When we interviewed mathematicians, one of the most common annoyances we heard about in their day to day work is battling with LaTeX compilation errors.
The most obvious solution is to ask a chatbot to fix errors. That works decently well, except switching to a different window breaks the writing flow, and copy-pasting the snippet often loses relevant context, especially when multiple files are involved.
So we scraped 272,000 questions from TeX.StackExchange and kept the 3,978 whose snippet still fails today, whose accepted answer compiles, and whose difference replays exactly as a patch. The largest dataset previously released had 88.
We tried tweaking the reward function in many ways. The first one was only a compilation check, which led the model to remove content until the document built. In the end, what worked best was rewarding compilation and a close match with the solution's PDF, and penalizing removed content.
We're rolling out the model inside the @sundialmd editor this week. When a compile fails, the fix is applied as a suggestion in under a second and the PDF rebuilt.
Write-up: sundial.md/blog/textinguishe…
Thanks @tinkerapi and @thinkymachines for the support.
Replying to @DennisonBertram
this is the kind of shit i want from agents
Nonplussed Netizen retweeted
Your Instinct can now handle phone calls.
Introducing Instinct Concierge – a white glove service meant to handle high-touch cases, such as making phone calls, high-end service booking, and more. You’ll be able to book the restaurant that doesn’t take online reservations, get on your dentist’s cancellation list, or have that cable bill sorted out.
We’re slowly rolling this out to our early access group today and will be expanding access soon.
The best intro to market making you can do is learning delta neutral funding rate arbitrage
It's possible to do manually and you can use screeners like loris.tools
So many 100%+ APY opportunities there and also much easier with small capital as you won't heavily flip rates on low OI markets
Funding rate opportunities are endless
Give it 6 months and we'll see @Cbb0fe post about how him and his brother made $10M from funding rate arbitrage by asking AI how to do it
I'm sure a lot of trading firms are milking this shit, it's not too late to learn how to build personal trading systems yourself
Bookmark this market making lesson from the QFEX founder @annanay
Replying to @MorganBarrettX @stephwakefield_
Nonplussed Netizen retweeted
DeepSeek kernel engineer:
Nonplussed Netizen retweeted
An alternative hypothesis:
- model performance is plateauing
- compute is getting much more expensive
- AI data centers are massively unpopular
- slowing down AI development is a way to explain slowing progress, reduce spending, and try to regain some goodwill.
- This is an effort to save the IPO not humanity.
We Must Pace the Frontier: I’ve written a new essay on why the AI industry should slow down, with a three-part plan for doing so.
Anthropic is unilaterally committing to the first of these steps. We’ll provide third-party evaluators with permanent, employee-level access to our systems, so that they can verify adherence to our safety measures, report on incidents, and assess models’ alignment during training.
You can read the full post here: darioamodei.com/post/we-must…
Replying to @DarioAmodei
All the regulations being floated by frontier model companies must be viewed through the lens of them losing tokens to Open-Source models.
The timing of these regulations seems to line up perfectly with Open Source closing the gap significantly (but not completely) — and @nvidia going all in on open source in the last 60 days.
If we’re gonna regulate, why don’t we require that last year’s frontier models and weights be open-sourced?
Replying to @DarioAmodei
This sort of regulatory capture will prevent homebrew companies like Slop Cannon from ever being allowed to exist. You think there will be third party AI evaluators available in Florida? in rural America? You think they'll be able to afford to red team their models to whatever standard Dario deems adequate?
No. All AI startups will only be allowed to use the vetted models from the big companies that can afford to vet them. And those models will be more expensive due to the vetting.
The reason America wins is because of those founders starting in their garage. Apple started in a garage. The spirit of American capitalism is in your garage.
And Dario wants to make sure that's illegal. That your opportunity is taken from you. That you are a custie, a customer, unable to partake in progress, paying extra for it, trapped.
He doesn't want intelligence to be a commodity, he wants YOU to be a commodity. He wants YOU regulated. He wants YOU stopped. He wants open source out of YOUR hands and he knows what it will take to ensure it.
That's what this is about. Stopping YOU. Stopping ME. Banning us from the fruits of civilizations progress that he pillaged from us. That he stole.
Fuck you, Dario.
Sincerely,
Bone