How users interact with our products in the real world is a valuable source for model training. The data is abundant, reflects the actual distribution of user tasks, and captures user corrections and tool failures that synthetic environments may miss.
A common way to learn from real-world data is rejection sampling fine-tuning: judge each session’s outcome, keep the successful ones, and train the model to imitate them. But a successful outcome does not mean every step was correct, so imitating the whole trajectory risks reinforcing bad intermediate behaviors in addition to good ones. Discarding unsuccessful sessions also loses critical evidence of where the model falls short.
We combine rejection sampling fine-tuning with hint-guided self-distillation to learn from both successful and unsuccessful sessions. A hint is a short corrective instruction grounded in information the model already had when it made the mistake. Useful steps from successful sessions remain imitation targets, while grounded hints turn avoidable mistakes into correction targets. In live use, the later trained checkpoint reduced tool-call failures by 21.2% relative to an earlier trained checkpoint. — Read More
Recent Updates
DeepSeek-V4.1-Flash: Frontier Agents on a Smaller Memory Budget
A lot of recent progress in language models has focused on better reasoning, larger reinforcement-learning runs, and increasingly sophisticated agent training. One of the many remaining bottlenecks is the context. Coding agents, research agents, and tool-using systems repeatedly process hundreds of thousands of tokens while generating comparatively little output. At that point, prefill compute and the KV cache become the most important infrastructure problems.
DeepSeek-V4.1-Flash is essentially an attempt to redesign the Transformer around that workload. Its 552B-parameter MoE backbone uses a new Causal Encoder-Decoder (CED) architecture, activating only about 8B parameters per token during prefill and 16B during decoding. Compressed Sparse Attention 2 (CSA2) shares KV representations and sparse-attention decisions across layers; SWA Bounded Replay avoids persistently storing sliding-window KV states; and FP4 KV quantization pushes the global cache down to only 890 bytes per token — about one quarter of DeepSeek-V4-Flash’s. The model still supports a 1M-token context, native vision, and strong agentic capabilities. It is trained on 45T multimodal tokens, with much of the improvement coming from scaling automatically generated agent tasks, environments, and rollouts. — Read More
Frontier Overhangs
There has been, over the last week, what I think is a healthy debate about the philosophy and psychology that undergirds the views of meaningful segments of the AI community, particularly those obsessed with doomsday scenarios. It is, in the end, difficult to reason with a philosophy that grants equivalent moral weight to not just all beings — human or not — who exist today, but who may ever exist in the future; this tilts the scales in such an absurd fashion towards safetyism that innovation is impossible and freedom is intolerable.
Worse, it taps into the psychology of religion, where dissent is not brooked and questioning the premise is heresy. I guess that makes me a heretic then: I reject the premise in favor of doubt in our ability to foresee the future, combined with faith in humanity figuring things out along the way. — Read More
dlab Open Source Week: Frontier AI on Your Own Hardware
In one of my classes I asked the question I was afraid to ask but I just needed the answer to: “Who is afraid of not getting a job after graduating?” About eighty percent of the 150 people in the room raised their hands. … The other story arrives by email. PhD students who cannot wait to graduate, because they want to join a frontier lab and they have concluded that research in academia is meaningless. They are counting the years until they can leave.
I believe both stories are wrong, and wrong for the same reason. They assume the future of research belongs to whoever has the most GPUs. I think the opposite is true. Academia is probably about to have a renaissance, and the most exciting work of the next decade will happen in university labs — not in spite of their limited resources, but because of them. — Read More
Andreessen Horowitz Wants Teens to Skip College for Its New Academy Instead
All of Silicon Valley seems to be coming together to launch a new alternative to college for teens dreaming of becoming the next big tech founder.
Venture capital firm Andreessen Horowitz, also known as a16z, announced Tuesday that applications are now open for the first class of the creatively named The Horowitz Andreessen Academy. — Read More
In His Voice
Through the VOICE trial, Terry is using his Neuralink implant to help fine-tune a brain-to-voice interface for himself and others who can’t speak. He trained the algorithm first by miming speech as best he could, then by simply thinking the words and hearing them come out in his own natural voice. — Read More
#audioAI Goes to War: Ukraine AI Autonomous Drone Guidance
Ukraine is making gigantic strides in AI autonomous drone guidance. There are two specific areas of focus:
— Last mile AI autonomy — a human pilot flies the drone to the target area and then AI takes over for the last 500 meters of flight guiding its warhead to exactly where it can do the most damage. BOOM!
— Mission AI autonomy — once parameters of the mission are set, an AI assisted drone flies without GPS/radio guidance to the target with no human involvement, Fire and forget. BOOM!
AI is the game changer. — Read More
Nvidia boss rejects AI extinction fears as ‘doomsday narratives’
Nvidia chief executive Jensen Huang has described warnings that AI could lead to humanity’s extinction by the next decade as “doomsday narratives”.
“2030 is not going to be the end of the world. There is 0% chance that’s going to be the end of the world”, Huang said in an interview with CBS News, the BBC’s US partner.
Huang added: “Scaring people is unnecessary. It is irresponsible.” — Read More
Human Brain is Two Organs Fused Together, Study Suggests
Scientists have long treated the brain as a single organ that grows from one pool of early cells. Stanford University neuroscientist Kyle Loh and colleagues challenge this view, arguing that the human brain is two distinct nervous systems that evolved separately and were joined together over hundreds of millions of years. The discovery could also help explain why researchers have struggled for decades to grow certain types of brain cells in a lab, and open new avenues for studying devastating diseases that affect the hindbrain, or brain stem, such as spinal muscular atrophy (SMA) and amyotrophic lateral sclerosis (also known as ALS or Lou Gehrig’s disease). — Read More
Read the Paper
‘Godfather of AI’ warns Congress has ‘maybe a year’ left to regulate AI
Geoffrey Hinton, a Nobel laureate known as the “Godfather of AI,” issued an ominous warning to lawmakers: Congress may have only one year left to implement safeguards on artificial intelligence before it loses control of it.
“Maybe a year, but not much more than a year,” Hinton said Wednesday evening after he and other AI experts huddled with Senate and House lawmakers behind closed doors. “If you look at predictions for when we’ll get superintelligence,” he continued, “it used to be maybe 30 years, maybe 50 years. Then it came down to maybe 10 years, maybe 20 years. Now people are saying, a lot of the researchers are saying only a few years.” — Read More
