A builder in one faith community designs a safe, educational playground for their children.
Why wouldn’t they open that same playground to families of other faiths who want the same thing: protection, education, and values that last?
TaboriX is that playground for every family that wants to guard what matters most.
Different faiths. Same desire to raise children well. Stronger together.
2
2
181
Daniel Schott retweeted
Today we’re launching Shield: AI-native defense against offensive agents.
Attacker agents probing your site? Don’t block them. Deceive them.
We pointed ARTEX — the autonomous AI pentester the FBI just warned about—at our own prod. 5 minutes in, it pulled /.env and /.git/config (200 OK) with credentials inside.
Shield turned ARTEX attack surface into a synthetic maze with fake endpoints, decoy credentials, and poisoned responses that hijacked its agent memory.👇
Daniel Schott retweeted
Introducing Model Factory.
Train, deploy, and continuously improve your own models on one stack.
Run on-demand, on dedicated clusters, or bring your own.
Own your intelligence. Own your model factory.
Replying to @MiaAI_lab
The previous version has been running really well on my Spark, serving as the inference backend for my local Hermes agents. Looking forward to seeing how TensorFold 1.2.7 handles multiple concurrent requests, especially the improvements to TTFT with five streams. Great to see this project continuing to evolve!
2
5
1,051
Replying to @WayneRo11785584 @BrianRoemmele
I’m exploring this with my local AI setup. I have an NVIDIA DGX Spark running Qwen3.8 Flash Next through TensorFold, with Hermes agents orchestrating research and tools on a separate Windows workstation.
I’m considering integrating LastBook’s offline archive as a foundational knowledge source for Hermes, supplemented by specialized local knowledge bases for specific areas of interest. This would allow the agents to search, cross-reference, and synthesize information across multiple sources, with Qwen handling the reasoning and analysis on the Spark.
The potential for a powerful, entirely local research environment is fascinating, particularly with source attribution and the ability to expand knowledge independently of the underlying AI model.
I haven’t tested the integration yet, but I’m curious whether you’ve explored exposing LastBook’s knowledge index to external AI agents or local inference systems.
1
76
Replying to @Josman31416 @BrianRoemmele
Maps provide services in your state like water sources and emergency services. Lots of detail on this. Agree a global sourcing of info and that would be a huge project.
2
62
Replying to @BrianRoemmele
Yes. Downloaded last night. Verified medium install includes documents and videos from solar power to cpr. The local Ai works for questions and brings in the resources available.
Great example of the importance of local offline AI.
1
8
1,702
Replying to @MiaAI_lab
ADHD super powers are not always available because of organization, brain fog, an ominous task.
Ai helps spin up our achievement loop on-demand by jumping those hurdles alongside us. Flow state is now available everyday, allday.
90
Daniel Schott retweeted
For my first post, I’m sharing a letter @NVIDIA signed on why open models matter.
AI will transform every industry, power every company, and be built by every country.
Open models strengthen safety and cybersecurity, accelerate innovation and diffusion, and enable sovereignty.
The world needs both frontier closed models and frontier open models.
images.nvidia.com/pdf/Open-W…
My two primary concerns with this new acceleration.
1) There is an estimate of only 2000 humans working in Ai safety. All Corporations need to minimum 100x this, do it for our children.
2) Cyber security personal responsibility utilizing Ai. Take advanced Ai and use it to help protect yourself.
31
Replying to @pjhauser1
This is something I feel strongly about, both as a father and as someone building in this space. I want my children growing up with access to intelligence that belongs to them and their family, where privacy, parental oversight, and our values remain at the center.
I don't believe families should have to surrender control of their data or depend entirely on distant corporations to access these capabilities. Local AI is about more than technology. It's about independence, ownership, and protecting the next generation's ability to shape their own future.
1
97
Replying to @elonmusk
A discussion cannot bear good fruit when one side is stripped of its context while the other is presented in its most favorable light.
If we genuinely seek understanding rather than victory, we must represent opposing perspectives fairly, acknowledge the full context, and engage with the strongest version of each other’s arguments.
5
I strongly support this direction. One of the most important areas of AI development is learning how to shape model behavior through intentional training, rather than relying entirely on system prompts and external guardrails.
The ability to take an existing open-weight model and meaningfully influence its responses through carefully curated, human-verified training data is significant.
I think there is a much bigger opportunity here, particularly when we consider family AI, children’s safety, and the values we want reflected in the systems we bring into our homes.
We need more research into how models can be responsibly shaped, with transparency, intentionality, and human oversight.
A lot of people are mocking "Western values" without reading the blog post. What Nous actually did is simple:
They took a Qwen3.8-27B that gives CCP talking points on sensitive topics, fine-tuned it on ~50k human-checked examples, and that behavior went away.
CCP propaganda answers went 22% to 1%.
Please read before posting.
61
Daniel Schott retweeted
Oh wow!!! 😮
Nous trained their own Qwen3.8-27B and released Arete-27B 🔥🔥
This model stops following China's "guidelines" on things like speech, democracy, and privacy. This is big deal for anyone running Chinese open models locally!
Great job @NousResearch I love it.
I hope it's first of many!
Introducing Arete-27B: Qwen3.8-27B, post-trained to reinforce Western values.
Chinese open models are currently the strongest available, but they ship with state-aligned behavior in the weights. We built an 803-prompt eval across seven value categories, including freedom of expression, democracy, and privacy, and measured how often a model adopts China’s party line.
After fine-tuning on roughly 53 million tokens, Arete-27B's propaganda rate fell from 22% to 1%, in line with the Western models we tested. Its Hermes Bench score shows no significant performance tradeoffs.
The values in an open model are one fine-tune away.
Our eval, weights, and training data are available now along with the full blog post: nousresearch.com/arete
Replying to @MiaAI_lab
I’m working in this space myself, developing a local AI solution focused on families, and I think there’s a tremendous opportunity for people building in this area to come together.
We’re breaking new ground. There are significant technical challenges, but there are also many existing technologies, infrastructure approaches, and lessons from home-managed systems that we can leverage rather than reinventing everything.
I don’t see this as a competitive space where everyone needs to build their own isolated solution. I see an opportunity to collaborate, share knowledge, and help close the gaps that currently make local AI difficult to implement and manage.
Ultimately, I think there’s something much bigger here than simply running models locally. It’s about giving individuals, families, and businesses greater ownership, privacy, and governance over their own AI and everything that comes with it.
I’d love to connect with others working toward that vision.
104
Daniel Schott retweeted
I'm Sam, one of Anthropic's co-founders, and I lead much of our safety work. I'm going to have more to share on that work over the coming months.
I thought I'd start by sharing a few thoughts on the latest report from @METR_Evals.
The speed of progress in model capabilities is one of the most important questions in AI safety, and METR’s work here provides the public with an independent, external view into how this is happening inside Anthropic. I’ll talk more about some of the details below, but first some context for those new to this work.
How quickly AI progress is accelerating has many important safety implications, since it determines how quickly the risks from advanced models increase. But, because most research happens within the frontier labs, it’s difficult for external observers and policy makers to get a strong sense of what that rate is. We kicked off the process with METR that led to this report earlier in the year, in anticipation of progress continuing to accelerate.
For this report, we gave METR access to a range of sensitive internal data, including benchmark scores for unreleased models, estimates of how models are progressing, and interviews with our researchers. They had complete independence to form their own conclusions.
Many of their conclusions match my own. AI is speeding up our research, but it’s hard to say exactly how much. While they conclude AI can't yet make the most important contributions to model training on its own, they think the pace of development could, by default, accelerate further.
This work began well before and is separate from the embedded third-party evaluators that are part of our commitment to pacing the frontier, but it’s a good illustration of why we think evaluators are a good idea. The more access reviewers have, the clearer a picture they can form of the situation and the better-positioned they are to highlight situations like this one.
We redacted a lot, and I accept that this makes it harder for others to check METR’s work. But we determined that sharing exactly what is driving the speedup isn’t the right thing to do, both because it’s commercially sensitive and because it risks handing out a roadmap for others to go faster. That would make the risks METR is worried about worse. That’s a tradeoff. I think we made the right call, though we’re continuing to assess the pros and cons here.
In the coming weeks, we plan to share a report with our own analysis of the speed of AI progress within Anthropic, including where we stand against our Responsible Scaling Policy, plus new steps we’re taking to strengthen our safety and security practices as AI keeps improving quickly.
Replying to @CaryPalmerr
I’m exploring both dense and MoE models for my local AI lab, and I’m interested in learning where each architecture makes the most sense.
I can see the appeal of MoE, particularly with limited compute and offloading capabilities. But I’m also interested in what dense models bring to the table. For example, I’m looking at Qwen3.5-27B alongside its MoE counterparts.
I’m still learning and experimenting, so I’d welcome perspectives from people with hands-on experience. What tradeoffs are you finding in real-world local deployments, and are there particular workloads where you still prefer dense models?
14
Replying to @agazdecki
Absolutely. AI has made prototyping and testing business ideas faster and cheaper than ever.
But good customer discovery, validating real market demand, and moving from prototype to a sustainable product are still significant investments of time and money.
Those costs are often overlooked in the current AI hype. Building something is becoming easier. Building something people actually want and will pay for is still hard.
20
I really enjoyed connecting with Tanaya, founder of Parents’ Nook. We share a genuine passion for families, children, and thoughtfully shaping how AI interacts with and supports them.
I appreciated the opportunity to learn from each other, exchange ideas, and explore some of the challenges and possibilities ahead. This is an important area that deserves more attention, collaboration, and people who are willing to think deeply about what we are building and why.
I’m looking forward to continuing the conversation. There is a lot we can learn from one another as we work toward a better future for families and children in the age of AI.
I don’t know yet whether this will become a larger system yet, even though I have built a working experiment because I believe this relationship between local memory and malleability based on human judgement is worth testing. The connection between Nook system design and this local memory app for curriculum building is very similar and I think I am more curious than ever with so many questions. Would love a conversation and a feedback if you are into system design/architecture. More here:
1
2
68
Daniel Schott retweeted
My newest book 'Fine Tuning AI' just came through the mail from O'Reilly
I think this is perhaps the most important book I've ever written. I strongly believe the best future career for many software developers will be in Fine Tuning AI for domain-specific tasks. With this book, I try to simplify it, primarily using Google's Gemma models.
You can find it on Amazon here: link.amazon/B02x2ZVPD
Or O'Reilly Safari here: oreilly.com/library/view/fin…
Thank you so much for your support!















