We previously highlighted @lightningrodai's data recipe for training forecasters. Their new work with @PTetlock and @VSatopaa adds another key piece: choosing the rule for scoring predictions, and how each one trades off accuracy and error.

Over the last few months at @river_ai_inc we’ve developed what we believe is the best system in the world for building your own internal custom models, driven by 10 years of experience training top models at DeepMind, OpenAI and xAI. In order to best support our customers, we’ve decided to do something radical: We also deploy our API on-prem on your compute cluster, and if you don’t have compute we will help you find it. Again and again we’ve seen that custom models designed for your own internal workflows outperform top frontier models on performance, at a fraction of the cost (1 percent or less). We’re just at the beginning of the custom model era, but its already clear that in the future more than 99% of LLM tokens will be generated on efficient custom models. The economics of it are just too powerful, and it’s far easier than people think once you have the right tools in place. The reason why small models will dominate is due to the physics of deep learning: A general model that can do everything needs more capacity and training it on the whole distribution of all tasks is more challenging, so the model will be larger and more expensive, and wont perform perfectly on every task. A small model trained on a targeted domain usually outperforms it.

Small bird, fast wings, Kolibri is here. 78B parameters. 3.46B active. Up to 1M tokens of context. Built in Europe. Now the weights are yours. Run it on your own hardware, under Apache 2.0.

Say hello to Echo, the best writing model at style imitation. Echo beats frontier models at writing tasks ranging from fiction to technical explanations, despite costing less than $5K to train.

New preprint from @lightningrodai, this time with Philip Tetlock and Ville Satopää! We post-train 5 versions of the same LLM, changing only the scoring rule used as the RL reward. Similar aggregate scores, but very different BIN profiles. A good Brier score alone doesn't tell you if a forecaster can distinguish likely from unlikely events. One might discern well but lose if its probabilities run systematically too high. A less discerning one might score better by hugging the base rate. BIN splits forecast performance into bias, information, and noise. Bias is a systematic shift in the probabilities. Noise is random scatter. Information is real signal about which outcomes are more likely — the part you need a powerful LLM for. Different uses call for different profiles. Reward choice is one lever shaping which forecaster you get. Congrats to co-authors @indiequant @KSkotheim64001 @VSatopaa @PTetlock 🙌 Full paper: arxiv.org/abs/2608.28482

We’re proud to share that @lightningrodai has been awarded grants totaling $1.6M from @coeff_giving and Longview Philanthropy! These grants will help strengthen Foresight AI forecasters for high-stakes decisions and build Model Factory: our platform that turns messy data into standardized RL environments for model training. Thank you for the support!

New preprint from @lightningrodai, this time with Philip Tetlock and Ville Satopää! We post-train 5 versions of the same LLM, changing only the scoring rule used as the RL reward. Similar aggregate scores, but very different BIN profiles. A good Brier score alone doesn't tell you if a forecaster can distinguish likely from unlikely events. One might discern well but lose if its probabilities run systematically too high. A less discerning one might score better by hugging the base rate. BIN splits forecast performance into bias, information, and noise. Bias is a systematic shift in the probabilities. Noise is random scatter. Information is real signal about which outcomes are more likely — the part you need a powerful LLM for. Different uses call for different profiles. Reward choice is one lever shaping which forecaster you get. Congrats to co-authors @indiequant @KSkotheim64001 @VSatopaa @PTetlock 🙌 Full paper: arxiv.org/abs/2608.28482

Now that the base open weights AI models are getting far better, and post training infra is becoming more mature and commercialized, there are going to be all new plays for companies that have large amounts of data to have their own models. Licensing data for external model training was previously the only play if you had a large corpus of information, but you can now reasonably go and train your own models as well without incurring the cost and complexity of competing with the labs on research. The general purpose frontier models will still have a leg up in broad areas due to the ability to handle the widest set of tasks, but you can absolutely see a future where we have far more models than today across every vertical and domain.

How do you train AI to predict the future, not just memorize the past? 🔮 In our latest episode, we explore how models master cause-and-effect reasoning to transform enterprise data into accurate predictions. Stream now: utm.io/usdRP

What models do the miners use? Most of the winners use closed source models (Haiku is at 44% on the top 50 vs 14% for the rest). This reverses for open weight models. For example Kimi K2.6 is the fifth most common model among miners but none of the top 50 miners use it. Foresight is an exception however, it is used across all miners, being the single most used model overall. Interestingly the top miner is almost the only one to use Gemini 3.1 pro.

Today, we are releasing Inkling-Small. Inkling-Small achieves comparable performance to Inkling at a quarter of its size. It features 276B total parameters, 12B active. We are making the full weights available. thinkingmachines.ai/news/ink… Fine-tune it on Tinker today, or chat with it in text, image, and audio on Tinker Playground.

One pattern I find useful for working with LLMs is a nice long ramble session. Sometimes the LLM needs more bits to understand what you're trying to achieve, but you're too lazy to type them. In these cases I like to lean back, switch to /voice and just ramble for like 10 minutes, total mess, anything goes, full stream of consciousness. Sometimes I declare it up top, something like "switching to speech recognition sorry for any typos...". Sometimes I turn it into a small interview of a few turns. But I find that the LLMs are somehow very good at reconstructing long incoherent rambles and often their echo of your own tangle of thoughts comes out quite a bit cleaner than what you started with. The result is that you improve the mind meld and have to correct things less from that point on.

Let's go! So important right now