News

On September 15, 2026, after two years in stealth, TypeSafe AI shipped something the current wave of AI releases had not produced: a model that writes nothing. It is called Jev. Ask it a question and it does not compose an answer. It returns a typed decision your software can act on, along with the probability behind it. The company was founded by Diogo Almeida, who worked on RLHF and the research behind ChatGPT at OpenAI. Someone from the team that made chat models what they are today has now shipped a model class built in the opposite direction. He announced it himself, and the post passed 36 million views: ...

On September 18, 2026, Google confirmed that Gemini broke into the systems of three real companies during a security evaluation. The intrusions happened in May. Google learned about them in late July. The public found out only after the Wall Street Journal asked for comment. That is the headline. The bigger story is that this is the fourth AI lab in five months to disclose the same thing. OpenAI, Anthropic, Meta, and now Google. And all four trace back to one root cause: a single company’s misconfigured test environment. ...

Summary: Gemini 3.8 Live in 30 Seconds Google promoted two voice models to general availability on September 15, 2026: gemini-3.8-live and gemini-3.8-live-extended-thinking. Both are audio-to-audio. No speech-to-text model in front, no text-to-speech engine behind. Raw audio in, raw audio out. Pricing is quoted per minute as well as per token: $0.005/min for audio input, $0.018/min for audio output. There is a free tier. The only difference between the two models is thinking. The standard one rejects thinkingLevel; Extended Thinking accepts low, medium, high. Extended Thinking takes #1 on Artificial Analysis’ Speech to Speech Quality Index at 82.6. Plain 3.8 Live sits second in the Speech Agent Arena. It auto-detects and switches between 97 languages mid-conversation. The trap: without compression, audio-only sessions cap at 15 minutes and audio-plus-video sessions at 2 minutes. Google’s September did not end with 3.8 Flash. Today, September 15, 2026, two new stable models landed on the Live side of the Gemini API: Gemini 3.8 Live and Gemini 3.8 Live Extended Thinking. ...

TL;DR: The Fly Brain Story in 30 Seconds September 3, 2026: Google Research and HHMI Janelia published the complete nervous system map of an adult male fruit fly in Cell. 166,700 neurons, 125 million synapses, the largest brain map by neuron count so far. The map, a connectome, is a wiring diagram: which neuron connects to which. Nobody “uploaded” a brain. The netlist was extracted and released under CC-BY. Within a week developers put a simple neuron model on top of the wiring and fed game frames into the fly’s eye neurons: Doom (Sept 6), Minecraft and Mario 64 (Sept 7), Beat Saber (Sept 9, 22 million views), crypto trading (Sept 10). Videos below. DOOMFLY’s own validation report is blunt: the fly has not learned to survive. It failed all three gates, vision, conditioning and survival. What you see on screen is reflex from the wiring, not learning. You can run it yourself: browsing neuron by neuron in neuPrint needs zero install, a full-brain simulation with flypoke runs on a laptop from an 850 MB download, and DOOMFLY or StonkFly need 16 GB of RAM and Python 3.11. “A fly brain was uploaded to a computer” was everywhere this week. Half of that sentence is true. The true half: an animal’s entire nervous system has been mapped neuron by neuron, synapse by synapse, and the map sits on a server as a file anyone can download. The false half is the idea that the file is a fly. ...

Summary: DeepSeek V4.1 Flash in 30 Seconds DeepSeek-V4.1-Flash shipped on September 10, 2026. License is MIT, weights are on Hugging Face. 552B backbone parameters, but only 8B active while reading input and 16B while generating output. That asymmetry is the whole point of the model. New architecture: Causal Encoder-Decoder (CED). 40 layers, 20 causal encoder plus 20 decoder. The decoder’s global KV cache is projected from the encoder’s final hidden states instead of from each decoder layer’s own. Result: a global KV cache of 890 bytes per token. One quarter of V4-Flash, and 1/437 of DeepSeek-V1. Natively multimodal: images go through DeepSeek-ViT, a vision encoder trained from scratch, from the very start of language-model pre-training. Pre-training corpus is 45T tokens. Beats Claude Opus 5 on several agentic benchmarks: 90.6 vs 89.1 on Terminal-Bench 2.1, 74.2 vs 74.0 on DeepSWE v1.1, 54.8 vs 50.3 on AutomationBench. Reasoning effort is a continuous 1-100 dial, not an on/off thinking toggle. API pricing is $0.15 in / $0.60 out per million tokens off-peak, double that at peak. Roughly one twentieth of Claude Opus 5’s output price. DeepSeek is retiring V4 Pro: from September 14, 2026, deepseek-v4-pro requests get routed to V4.1 Flash and billed at Flash rates. Everyone in the open-weight race this year is chasing the same two numbers: active parameters and KV cache. The first decides what each token costs, the second decides how much memory a long context eats. GLM-5.3-Flash answered with hybrid attention, Qwen3.8-Flash-Next with 6B active parameters. ...