Tenago Labs

?? Tenago Labs won first place at the Inland Northwest Artificial Intelligence Hackathon for our dynamic illustration engine that helps early readers engage more advanced books by creating visual anchors as mental context for dense material. Functional overview: ?? Read in an eBook object, building contextually accurate images for each dynamically defined "page"; multi-paragraph page definitions support more dense material, like The Hobbit, whereas single or double paragraph pages provide better visual cues for faster moving material like chapter books. ?? Create engaging visual anchors to help young readers connect with dense material that requires a longer mental context window. ?? AI inference based on open source Mistral AI Mistral-Nemo (4bit local) for LLM processing and open source Black Forest Labs Flux.1-Schnell (hosted) for image generation. Architecture: - Three layers of recursive language processing to generate final image prompts (local Mistral-Nemo model inference): 1. Running context maintains thematic continuity; 2. Local visual cues provide environmental context; 3. Reader tracking object generates images specific to the particular story element as user reads the text aloud. - Aggregated responses are combined to create a rich image generation prompt that maintains visual consistency and contextual accuracy (hosted Flux.1-Schnell model inference). - Pipeline code repo: https://lnkd.in/g34SvMXm ?????? Huge THANK YOU ?????? ?? Hugging Face (Open source model hosting and test spaces) ?? Ollama (Low overhead local inference) ?? LangChain (Rapid LLM development pipeline) This project would never have been possible without the generous contributions of Hugging Face, Ollama, and LangChain projects to the Open Source AI community - decentralized AI is the foundation of a benevolent and equitable AI future, and I truly appreciate the work of these amazing groups and the life it gives to innovators around the world. -------------------- Check out a few of the other amazing projects emerging from Inland Northwest's premier Artificial Intelligence Hackathon and AI Think Tank: ?? My Protege: Share your knowledge at scale with a digital protege (https://www.myprotege.ai/) ?? Dispatch Robot: AI powered freight assistant that optimally matches trucks and loads in real time. (https://lnkd.in/gCKN_53g) ?? Code Canvas Generator: Visually explore complex or unknown codebases. (https://lnkd.in/gvRFqnz4) ?? Homoiconic LLM AGI: Leveraging weights as input and output to provide general intelligence reasoning layer on top of traditional LLM models. (https://lnkd.in/gMvJtD9Z) #OpensourceAI #AIFreedom #OwnYourIntelligence

Article on Idaho's "Freedom of Intelligence" - thoughts?

"The Microsoft team’s new measurement approach is so precise it can detect the difference between one billion and one billion and one electrons in a superconducting wire – which tells the computer what state the qubit is in and forms the basis for quantum computation." Between Microsoft Azure's quantum progress and Google DeepMind's Titan architecture, we will see galactic Von Neumann constructors within a decade; SpaceX - I hope your Starship will be ready to take it to Mars! More details: "Our solution to the [hidden qubit] measurement challenge works as follows: 1. We use digital switches to couple both ends of the nanowire to a quantum dot, which is a tiny semiconductor device that can store electrical charge. 2. This connection increases the dot’s ability to hold charge. Crucially, the exact increase depends on the parity of the nanowire. 3. We measure this change using microwaves. The dot’s ability to hold charge determines how the microwaves reflect off the quantum dot. As a result, they return carrying an imprint of the nanowire’s quantum state. We designed our devices so these changes are large enough to measure reliably in a single shot. Our initial measurements had an error probability of 1%, and we’ve identified clear paths to significantly reduce this." Thank you Chetan Nayak for the amazing research and insightful summary!

AI 2.0 Industry Guide - Spot on current-to-future state mapping for 2025. I especially appreciate the "unimaginative products" transition,"defined global frameworks", "federated learning", "focus on augmentation, and "engineered curiosity" elements. Thank you Stephen Klein for putting together this useful and insightful resource - it is a worthwhile roadmap and provides a solid foundation for resource allocation throughout the year.

MIT and Cornell agree that one can “Durably reduce conspiracy beliefs through dialogues with AI”.?The study allowed a three-step dialog with ChatGPT-4 Turbo, with instructions to reduce the user’s belief in the study topic.?As a result, “[t]he treatment reduced participants’ belief in their chosen conspiracy theory by 20% on average. This effect persisted undiminished for at least 2 months”. How can this simple exchange undo decades of personal investment and deeply held, emotionally significant, conviction? -??????Is it the natural outcome of intentional conversation with an adversarial viewpoint? -??????Personalized voice inherent in LLM text completion architecture? -??????The model’s hard coded veil of empathetic humility? -??????Perception of machine authority and neutrality? It is interesting to note that the questions were preconditioned as “conspiratorial”, with instruction to “reduce conspiratorial perspective” - No pretense of truth, exploration, or mutual understanding, and yet belief in this deeply held, personally significant, thoroughly researched perspective was reduced by 20% in a single three-shot dialog. The mere instruction to reduce belief in “X” was sufficient to accomplish the task in a significant portion of the target population. What does this conversational super power mean for the future of truth??How long would it take to establish an alternate worldview through mass adoption of an Alternate Intelligence? He who holds the weights, holds the future? Is there an antidote? Do we care? Thank you, MIT Sloan School of Management and Cornell University for the insightful and thought provoking research.

The U.S. Copyright Office recently released their current guidance on how AI fits into current Copyright laws and doctrine. TLDR: AI technologies fit well into existing legal and doctrinal frameworks; “generated” content is treated as co-authored or derivative material, but creators can protect those specific expressive elements which are conserved through the generative process.?Prompts are generally considered instructional ideas, which are not protectable. ? Overview of potential foundations for copyright protection within the Generative AI creation process: ? -??????Prompting: prompts lack sufficiently tight control over output for them to be a form of authorship; the indirect translation, system level augmentation of the prompts, and uncontrolled medium transfer make prompting an insufficient basis for copyright protections.?This position is reinforced by the observation that only ideas, rather than created material, persist from the prompt into the final product.?Prompts remain in the realm of instruction and ideas, which are not considered protectable personal expression. ? -??????Expressive Inputs: expressive guiding contributions (ex. sketches, musical samples, embedded datasets) only confer authorship to the extent that these elements identifiably persist into the final product; to the extent they persist, they are only incrementally protected, similar to protections for new contributions to a derivative work.?The elements of the expressive input which show up in the output are still protected, but the additive elements contributed by the AI element are not protected, since they are analogous to the pre-existing elements of a derivative work. ? -??????Modifying or Arranging AI Generated Content: sufficient modification of AI output may qualify the final work as a whole qualifies for protection as an original work of authorship.?The individual AI generated components are not protected, but if they are arranged or modified (ex. edited with AI infill / remix tools, images combined with text, AI augmentation, or mixed media material assembled into a comic book), then the aggregate qualifies for authorship protections. ? What do you think of these findings??How do these concepts intersect with AI platform claims to ownership of, and right to use, consumer output? Does it change how you feel about sending data and ideas to public AI service? If you would like to look at solutions to protect your ideas and data assets, let’s book some time to chat! Reference: https://lnkd.in/gW5Rhaya

Even if it is a distillation, it begs the question of whether re-compressing an existing model without permission is materially different from re-compressing the internet without permission; this is a door I'm not certain OpenAI wants to push too hard. On the technical front, the significant question would be "how" the knowledge was extracted (if, indeed it was) from ChatGPT. If it came through public endpoints as a synthetic generation process, it is not "distillation", since the API does not provide the token probabilities which one would need for the back propagation cycle; if it is a true distillation, then the builders would have had to access the source code directly, to expose these parameters in an expatriated offline context. At the end of the day, it would not be possible to "legally distill" OpenAI models because: A: One is hitting a public endpoint without the necessary propagation parameters (ie. it's not distillation), or B: One is hitting a modified pirate endpoint with the necessary propagation parameters (ie. it's not legal).

If the DeepSeek release does nothing else, it puts to final rest, the Altman, Musk, Biden, EU, and California argument that tightly regulated AI in the hands of a few trusted agents is the only sustainable, safe, and viable path forward. Thank you, DeepSeek for exposing this naked emperor for what it is. Mark Zuckerberg and AI at Meta - You called it from day one. Now it's time to quit fretting and freaking, so that we can get started bringing together these innovative memory techniques with our unique datasets, perspectives, use cases, and compute resources to jump start the next era of distributed intelligence.

I wonder if this is an example of innovative disruption, or more an example of how shallow is the AI moat; if DeepSeek came first, I would lean toward to the former, but given its position as a late follower, even after other open source platforms (Mistral, Qwen, Llama), the latter seems equally or more likely. Leveraging existing open source models, potentially expatriated resources, and commercial model access to create a roughly equivalent product at a fraction of the cost isn't as earth shattering as it is revealing.

"Autonomy is the Differentiator" - true [somewhat]; autonomy certainly separates agentic workflows, but at some level, this is an incremental improvement over existing automation layers. In Tenago's approach, composability and adaptability provide a much more resilient differentiator; traditional automation tools require deep code stack access, precise data mapping, and highly static environments. What separates Tenago's AI agent flows from traditional automation is their adaptive composability - the ability to couple: - a time series AI monitoring environmental factors to - a visual model identifying contextual activities to - an operational agent which can navigate an existing SCADA UI with no API access to - an image analysis model which can extract amorphously arranged values from a screen to - a natural language processor which adds context from an externally managed database to - an image generation agent which provides a meaningful graphic based on another agent's evolving knowledge of the user's preference. There is no operational model in which this flow, or the desired outcome, could emerge from traditional SaaS products; any reasonable approximation would require NDA's, modern instrumentation, API developers, impossibly static endpoints, etc... and would still break whenever someone failed to hold their tongue just right when clicking "build". If you are ready to extract intelligence without handcuffing yourself to another SaaS, please drop us a line!