OpenAI has fired an employee over the employee’s activity on prediction markets, including Polymarket, the company confirmed to Wired. The employee used confidential OpenAI information in connection with the trades made, the company alleges.

OpenAI didn’t release the name of the employee. However, a spokesperson said that such actions violated a company policy that bans workers from using inside information for personal gain, including on prediction markets.

Prediction markets like Polymarket and Kalshi allow people to make wagers on the outcomes of real-world events. For instance, on Polymarket, there are wagers being made around the kind of products OpenAI will announce in 2026 and when the company will go public. They can cover any event, and some eye-popping money can be made. As we recently reported, an accountant won a $470,300 jackpot on Kalshi by betting against DOGE believers.

Prediction markets insist they are not gambling sites, preferring to label themselves as financial platforms. Kalshi is a regulated exchange and, in fact, it fined and banned a MrBeast editor for similar alleged insider trading earlier this week. OpenAI did not immediately respond to a request for additional comment.

Photo credit: Onocom
O House, finished in late 2025 within Kurihara, Miyagi Prefecture, is a 50-square-meter, 3D-printed two-story home. The 31-square-meter ground floor holds a master bedroom and bathroom, while the 19-square-meter upper level hosts the kitchen, dining, and living areas. Curved walls rise 7 meters, stacked like bricks and set half a meter below ground for stability, while skylights let natural light pour into every corner.

Kizuki Co. Ltd. and Onocom Co. Ltd. jointly led this project, while COBOD International handled the printing technology. A modified version of their BOD series printers was exactly what they needed to get the job started. Working layer by layer, the printer built the inner and exterior walls, floor slab, roof, and a few inside elements. The printer handled on-site work, however some components were manufactured off-site. To cap it off, the crew installed custom-cut styrofoam for those problematic overhangs and arches that the printer couldn’t handle and increased as the process progressed. Finally, the team gave some of the wall parts a shining polish to get the smooth marble appearance.

Bambu Lab A1 3D Printer, Support Multi-Color 3D Printing, High Speed & Precision, Full-Auto Calibration…

• High-Speed Precision: Experience unparalleled speed and precision with the Bambu Lab A1 3D Printer. With an impressive acceleration of 10,000 mm/s…
• Multi-Color Printing with AMS lite: Unlock your creativity with vibrant and multi-colored 3D prints. The Bambu Lab A1 3D printers make multi-color…
• Full-Auto Calibration: Say goodbye to manual calibration hassles. The A1 3D printer takes care of all the calibration processes automatically…

They used a combination of rebar inserted directly inside the layers of concrete and good old-fashioned reinforced steel, all linked together by a steel frame that bears the primary loads. With a firm anchor in the ground thanks to some ground-improvement piles, this hybrid technology enabled the construction to exceed Japan’s stringent seismic regulations, which are among the strictest in the world. All of this demonstrates that in earthquake-prone regions, printed reinforced concrete is a viable alternative to timber framing.

Rounding up a crew of only four in challenging winter conditions made things tough, with freezing temperatures below 10 degrees demanding heated water be added to the mix to keep things flowing, while hot summers of 30 to 35 degrees required careful temperature control to stop the material from setting too fast, but they kept going, working continuously from below ground to the top, and creating multi-purpose walls with an aesthetic finish, structure, and hidden services.
[Source]

BMW employees at the Leipzig plant have been juggling all the complex parts of vehicle assembly, especially the hefty battery modules. However, a new member has joined the team: the AEON humanoid robot. AEON was created by Hexagon Robotics, a company BMW has been collaborating with for years on grunt work, to manage the physically taxing and repetitive tasks that wear people out.

By April 2026, the AEON prototype will be put through its paces in a larger round of assessments, with a complete pilot phase scheduled to begin this summer. The goal is to make AEON extremely adaptable so that it can transition between activities as needed; simply switch out the gripper or scanner and you’re ready to go. It travels from station to station without the need for fixed rails because it has wheels rather than legs.

Unitree G1 Humanoid Robot（No Secondary Development）

• Height, width and thickness (standing): 1270x450x200mm Height, width and thickness (folded): 690x450x300mm Weight with battery: approx. 35kg
• Total freedom (joint motor): 23 Freedom of one leg: 6 Waist Freedom: 1 Freedom of one arm: 5
• Maximum knee torque: 90N.m Maximum arm load: 2kg Calf + thigh length: 0.6m Arm arm span: approx. 0.45m Extra large joint movement space Lumbar Z-axis…

Electric vehicle battery modules need extra care, and workers frequently need to wear safety gear simply to move them. After a few shifts, it becomes monotonous, but AEON is more than willing to relieve them of that kind of work without putting undue strain on the human workers. External component production is also relevant since, let’s face it, robotic consistency is a huge bonus when performing the same operation repeatedly.

BMW is doing something a little different from the conventional industrial arms that are fastened to the ground. Thanks to data from BMW’s recently unified systems, AEON is able to move around, adjust to any arrangement, and become more intelligent every day. For a more seamless operation, the business had to dismantle the outdated data silos that were creating so much friction; now, all of that data streams directly into AEON. Naturally, safety is the top priority, which entails improved wireless coverage, additional barriers to keep people safe, and other measures. Since everything is linked into the current Smart Robotics network, there is no need to worry about anything getting left behind.

The main benefit in this case is that human labor is still essential to the entire operation. BMW wants employees to be able to do something more interesting for a change by eliminating the monotony of their jobs. The personnel on the floor have been won over by early buy-in from safety teams, IT, and logistics, and having their support from the beginning has made everything go much more smoothly. The strategy here was shaped by lessons learned from an earlier test in the US facility in Spartanburg, which demonstrated how rapidly AEON could catch up and the dependability of everything in an actual production environment.

In all of this, Leipzig has established a new Center of Competence for Physical AI, and its experts are getting to work assessing partners, conducting pilots, and expanding the concepts that prove effective. To stay competitive in Europe, executives are already discussing quantifiable improvements in speed and accuracy for these difficult activities.
[Source]

In building LLM applications, enterprises often have to create very long system prompts to adjust the model’s behavior for their applications. These prompts contain company knowledge, preferences, and application-specific instructions. At enterprise scale, these contexts can push inference latency past acceptable thresholds and drive per-query costs up significantly. 

On-Policy Context Distillation (OPCD), a new training framework proposed by researchers at Microsoft, helps bake the knowledge and preferences of applications directly into a model. OPCD uses the model’s own responses during training, which avoids some of the pitfalls of other training techniques. This improves the abilities of models for bespoke applications while preserving their general capabilities. 

Why long system prompts become a liability

In-context learning allows developers to update a model’s behavior at inference time without modifying its underlying parameters. Updating parameters is typically a slow and expensive process. However, in-context knowledge is transient. This knowledge does not carry across different conversations with the model, meaning you have to feed the model the exact same massive set of instructions or documents every time. For an enterprise application, this might mean repeatedly pasting company policies, customer tickets, or dense technical manuals into the prompt. This eventually slows down the model, drives up costs, and can confuse the system.

“Enterprises often use long system prompts to enforce safety constraints (e.g., hate speech detection) or to provide domain-specific expertise (e.g., medical knowledge),” said Tianzhu Ye, co-author of the paper and researcher at Microsoft Research Asia, in comments provided to VentureBeat. “However, lengthy prompts significantly increase computational overhead and latency at inference time.”

The main idea behind context distillation is to train a model to internalize the information that you repeatedly insert into the context. Like other distillation techniques, it follows a teacher-student paradigm. The teacher is an AI model that receives the massive, detailed prompt. Because it has all the instructions and reference documents, it generates highly tailored responses. The student is a model being trained that only sees the main question and doesn’t have access to the full context. Its goal is simply to observe the teacher’s responses and learn to mimic its behavior.

Through this training process, the student model effectively compresses the complex instructions from the teacher’s prompt directly into its parameters. For an enterprise, the primary value happens at inference time. Because the student model has internalized the context, you can deploy it in your application without needing to paste in the lengthy instructions again. This makes the model significantly faster and with far less computational overhead.

However, classic context distillation relies on a flawed training method called “off-policy training,” where the model is trained on fixed datasets that were collected before the training process. This is problematic in several ways. During training, the student is only exposed to ground-truth data and teacher-generated answers, creating what Ye calls “exposure bias.” In production, the model must come up with its own token sequences to reach those answers. Because it never practiced making its own decisions or recovering from its own mistakes during training, it can easily derail when operating independently. It’s like showing a student videos of a professional driver and expecting them to learn driving without trial and error.

Another problem is the “forward Kullback-Leibler (KL) divergence” minimization measure used to train the model. Under this method, the model is graded on how similar its answers are to the teacher, which encourages “mode-covering” behavior, Ye says. The student model is often smaller or lacks the rich context the teacher had, meaning it simply lacks the capacity to perfectly replicate the teacher’s complex reasoning. Because the student is forced to try and cover all those possibilities anyway, its underlying guesses become overly broad and unfocused.

In real-world applications, this can result in hallucinations, where the AI gets confused and confidently makes things up because it is trying to mimic a depth of knowledge it does not actually possess. It also means that the model cannot generalize well to new tasks.

How OPCD fixes the teacher-student problem

To fix the critical issues with the old teacher-student dynamic, the Microsoft researchers introduced On-Policy Context Distillation (OPCD). The most important shift in OPCD is that the student model learns from its own generation trajectories as opposed to a static dataset (which is why it is called “on-policy”). Instead of passively studying a dataset of the teacher’s perfect outputs, the student is given a task without seeing the massive instruction prompt and has to generate an answer entirely on its own.

As the student generates its answer, the teacher acts as a live instructor. The teacher has access to the full, customized prompt and evaluates the student’s output. At every step along the student’s generation, the system compares the student’s token distribution against what the context-aware teacher would do.

OPCD uses “reverse KL divergence” to grade the student. “By minimizing reverse KL divergence, it promotes ‘mode-seeking’ behavior. It focuses on high-probability regions of the student’s distribution,” Ye said. “It suppresses tokens that the student considers unlikely, even if the teacher’s belief assigned them high probability. This alignment helps the student correct its own mistakes and avoid the broad, hallucinatory distributions of standard distillation.”

Because the student model actively practices making its own decisions and learns to correct its own mistakes during training, it behaves more reliably when deployed in a live application. It successfully bakes complex business rules, safety constraints, or specialized knowledge directly into its permanent memory.

What OPCD delivers: The benchmark results

The researchers tested OPCD in two key areas: experiential knowledge distillation and system prompt distillation. For experiential knowledge distillation, the researchers wanted to see if an LLM could learn from its own past successes and permanently adopt those lessons. They tested this on models of various sizes, using mathematical reasoning problems.

First, the model solved problems and was asked to write down general rules it learned from its successes. Then, using OPCD, they baked those written lessons directly into the model’s parameters. The results showed that the models improved dramatically without needing the learned experience pasted into their prompts anymore. On complex math problems, an 8-billion-parameter model improved from a 75.0% baseline to 80.9%. For example, on the Frozen Lake navigation game, a small 1.7-billion parameter model initially had a success rate of 6.3%. After OPCD baked in the learned experience, its accuracy jumped to 38.3%.

The second set of experiments were on long system prompts. Enterprises often use massive system prompts to enforce strict behavioral guidelines, like maintaining a professional tone, ensuring medical accuracy, or filtering out toxic language. The researchers tested whether OPCD could permanently bake these dense behavioral rules into the models so they would not have to be sent with every single user query. Their experiments show that OPCD successfully internalized these complex rules and massively boosted performance. When testing a 3-billion parameter Llama model on safety and toxicity classification, the base model scored 30.7%. After using OPCD to internalize the safety prompt, its accuracy spiked to 83.1%. On medical question answering, the same model improved from 59.4% to 76.3%.

One of the key challenges of fine-tuning models is catastrophic forgetting, where the model becomes too focused on the fine-tune task and worse at general tasks. The researchers tracked out-of-distribution performance to test for this tunnel vision. When they distilled strict safety rules into a model, they immediately tested its ability to answer unrelated medical questions. OPCD successfully maintained the model’s general medical knowledge, outperforming the old off-policy methods by approximately 4 percentage points. It specialized without losing its broader intelligence.

Where OPCD fits — and where it doesn’t

While OPCD is a powerful tool for internalizing static knowledge and complex rules, it does not replace all external context methods. “RAG is better when the required information is highly dynamic or involves a massive, frequently updated external database that cannot be compressed into model weights,” Ye said.

For enterprise teams evaluating their pipelines, adopting OPCD does not require overhauling existing systems or investing in specialized hardware. “OPCD can be integrated into existing workflows with very little friction,” Ye said. “Any team already running standard RLVR [Reinforcement Learning from Verifiable Rewards] pipelines can adopt OPCD without major architectural changes.”

In practice, the student model acts as the policy model performing rollouts, while the frozen teacher model serves as a reference providing logits. The hardware requirements are highly accessible. According to Ye, enterprise teams can reproduce the researchers’ experiments using about eight A100 GPUs.

The data requirements are similarly lightweight. For experiential knowledge distillation, developers only need around 30 seed examples to generate solution traces. Because the technique is applied to previously unoptimized environments, even a small amount of data yields the majority of the performance improvement. For system prompt distillation, existing optimized prompts and standard task datasets are sufficient.

The researchers built their own implementation on verl, an open-source RLVR codebase, proving that the technique fits cleanly within conventional reinforcement learning frameworks. They plan to release their implementation as open source following internal reviews.

The self-improving model: What comes next

Looking ahead, OPCD paves the way for genuinely self-improving models that continuously adapt to bespoke enterprise environments. Once deployed, a model can extract lessons from real-world interactions and use OPCD to progressively internalize those characteristics without requiring manual supervision or data annotation from model trainers.

“This represents a fundamental paradigm shift in model improvement: the core improvements to the model would move from training time to test time,” Ye said. “Using the model—and allowing it to gather experience—would become the primary driver of its advancement.”