One hope for Tinker is that it can handle all the ML infra for posttraining for (approximately) everyone. So, curious, what prevents you from using Tinker for all your posttraining needs? Speed? Cost? Scalability? Paranoia about correctness? Flexibility? Let me know!

posting articles is fun. so far, tons of outreach for: - access to the AI Cloud TCO model (which includes our Neocloud Dashboard, below) for pricing data, site maps, company financials, input costs, and more - providers getting in for ClusterMAX 3.1, including some big GB300 deployments that we missed testing - customers with some crazy stories about reliability, security, and performance - investors going down the list to find the next big Neocloud that can address this mid-market need for managed clusters for the Neolabs - access to sample SLAs, and intros to insurance providers who are backing these rental contracts - participation in PostTrainingX, where we test managed RL infrastructure services and RLaaS/FDE work (or as they call it, forward deployed researchers - participation in SandboX (and to review initial results) where we benchmark CPU/memory setups for agentic coding environments - participation in EndpointX (and to review initial results) where we test managed inference endpoints - security research and a ton of good applications to join the team. please reach out!

GOOGLE SET TO RELEASE GEMINI 3.8 FLASH THIS WEEK: WSJ Google $GOOGL is expected to release Gemini 3.8 Flash, internally called “Skimaki,” as soon as Wednesday, with a major focus on coding. In internal head-to-head tests inside Google’s coding tool Jetski, engineers reportedly preferred 3.8 Flash over Anthropic’s Opus model. Flash is designed to be smaller, cheaper and faster than Google’s flagship Pro models. Google has also scrapped internal Gemini 3.5 Pro candidates that weren’t sufficiently better than Flash, while Gemini 4 has performed well in pretraining evaluations but still needs to complete posttraining.

📺 Xiaomi Is Livestreaming a Production RL Run — Burn Rate Included @XiaomiMiMo is training its MiMo-V2.6 in public — not a demo, a live post-training dashboard where every step's reward, loss and benchmark movement is watchable in real time. One day in, the meter reads roughly $1.2M spent and 60B tokens consumed. As Zhihu contributor Kitt在进化 puts it: this isn't a training site, it's a money-burning site — about ¥4,000 a minute. His bigger point: this is rare, valuable data. Almost nobody publishes what frontier-scale RL actually costs. 1️⃣ What the dashboard actually shows Every training step exposes the internals most labs keep private. His reading guide: 🔹 Loss family: training loss, entropy (are answers diversifying or collapsing into one mode), gradient norm, and train/inference KL divergence. 🔹 Reward: a mix of test-case scoring and rubric-based reward. 🔹 dynsam (dynamic sampling): avg@n performance over repeated tries, plus pass-rate buckets — including samples never solved and always solved. 🔹 DeepSWE v1.1 as the headline probe (mini-swe-agent, avg@3): Pro at 63.7, Flash at 60.7 at the time of writing. 2️⃣ The cost ledger, out in the open From the dashboard he extracts the numbers the industry usually guesses at: 🔹 MiMo-V2.6 Pro: roughly $36 per million tokens of training. 🔹 MiMo-V2.6 Flash: roughly $8 per million tokens. 🔹 His rule of thumb: equivalent inference runs 30-50x cheaper than these training figures. 3️⃣ What's actually being trained The sample distribution is also public: about 1,500 prompts per step, ~70% of them coding tasks. The remaining third is split across general, visual, cybersecurity and chat tasks — which makes this, in effect, an agentic-coding-centric RL run with side dishes. 4️⃣ Why it's worth watching He notes a university lab livestreamed a training run days earlier, but at nowhere near this scale or frontier relevance. For anyone who wants to learn how production RL behaves — rewards, entropy, dynamic sampling, benchmarks moving step by step — this is a rare open classroom. His only complaint, half-joking: a livestream this good deserves a comment section. 🔗 Full Reading: #Xiaomi# #MiMo# #ReinforcementLearning# #PostTraining# #LLM# #OpenScience# #AI#

TL;DR: "The model you use is not the model that was pretrained." An 87-slide lecture that maps post-training — turning a base model into a useful policy — across six stages: imitate, compare, explore, verify, transfer, anticipate. Title: Post-Training LLMs (Kawin Ethayarajh, AI and Economics Summer Institute 2026) URL: Key points 📝 SFT: teach by demonstration. Can imitate whole trajectories (plan, tools), but it's fundamentally imitative ⚖️ Offline preference optimization: learn from win/loss. DPO drops the reward model; KTO learns per-outcome, no SFT needed 🎲 Online RL: the current policy explores. REINFORCE/PPO/GRPO; rewards are mostly sequence-level ✅ RLVR: replace the reward model with a checker. Binary pass/fail is enough; scales more predictably than RLHF 🧪 Environments: an RL env is task + data + interface + tests. The bottleneck shifts from examples to environments 🔁 On-policy distillation: the teacher scores the student's own rollout. Copies an RL policy in 7–10x fewer steps 🌐 World adaptation: environments adapt back to agents. "Mecha-nudges" raise machine-readability A systematic map, threaded with an economist's lens. #LLM# #PostTraining#

Post-training is where frontier models get sharper - and where compute gets expensive. At Exploit, @0xgrizzz + @RomAI__ of @reliquary_ai show how SN81 turns reinforcement learning into an open market: miners hunt for training signal, prove the work + compete to improve the next checkpoint:

Does post-training actually create new capability, or does it just sharpen what was already there? Title: Sharpening Tax in Post-Training URL: ❓ What was the surprising finding? A pretrained base model, given nothing but a lightweight harness, can handle agentic tasks. Its single-shot accuracy is far worse, but once you let it retry, it overtakes its post-trained counterpart on solution coverage. On WebShop with gemma-4-31B, the base model passes over 85% at pass@128 while the RL version sits at 56%. 💡 Why does that happen? Post-training pushes tasks toward two extremes, always solved or never solved. In the same setting, the middle band where extra attempts still convert failures into successes collapses from 87.6% to 30.0%, while never-solved grows from 12.4% to 44.0%. There is simply no headroom left for more rollouts to recover. ❓ What is the Sharpening Tax? A single-number diagnostic for how much test-time scalability was lost relative to the base model. It comes out positive in 36 of 42 model-benchmark combinations, and it grows with model scale. It's also cheap to estimate: the value from just 8 rollouts predicts the tax at 32 rollouts with a Spearman correlation of 0.85. 💡 Can it be avoided? Not entirely, but it can be reduced. The proposed PTGS estimates each prompt's difficulty with a Bayesian posterior, then heats hard prompts to encourage exploration and cools easy ones to enforce consistency. It touches neither the RL loss nor the compute budget, and improved accuracy and coverage together under both PPO and GRPO. #ReinforcementLearning# #LLM#

AlphaSense is post-training models end-to-end for search and research per SVP of Product @cackerso. The research loop is 80% of cost and 70% of quality in their agents, and AlphaSense expects its specialized research model can cut costs 40x.

“Why Does Post-Training Quantization Work?” This paper shows that pretrained LLMs are inherently robust to quantization because layers actively counteract accumulated quantization errors, while high-dimensional LM-head geometry preferentially preserves top-token predictions. This lets 4-bit models stay close to full precision despite substantial hidden-state perturbations.