Zelig retweeted
New Anthropic research: progress on provable inference for large language models.
Here, we introduce FlashProve. Building on established cryptographic techniques, FlashProve allows developers to verify whether a given response came from a given AI model. It’s substantially more efficient than prior open-source methods for doing so.
These proofs have a range of applications. They can, for instance, make it harder for attackers to hijack a chat, or enable safety evaluators to verify that the model they’re testing is the same one being served to customers.
For our full write-up, see here: anthropic.com/research/prova…
Further progress in provable inference will likely be an important part of improving trust in AI deployment. To that end, we're open-sourcing FlashProve’s code, here: github.com/anthropics/FlashP…
We’re grateful to our collaborators: Alessandro Chiesa, @GiacomoFenzi, @gwberg95, @miranda__christ, Rahul Dalal (rdalal.github.io) and Ben Mayer (@transmissions11).
Zelig retweeted
MiniMax M3.1 Flash: Space Performance + Bunny Speed
🚀🐰 Yes, Space Bunny - which has topped global leaderboards with 200T+ tokens - was an earlier version of M3.1 Flash.
💰Redefining the performance economics of agentic AI. M3.1 Flash achieves frontier-level intelligence across key tasks such as Cowork (Automation-Bench: 62%; ALE: 30%), Coding (DeepSWE: 74%) and Visual Reasoning (BabyVision: 91%), at 1/3 the cost of M3 and 3x the speed.
API pricing (flat rate): $0.15 / $0.60 per 1M input/output tokens, with cache reads at $0.015 / 1M tokens.
Model weights will arrive on Oct 23.
API: platform.minimax.io
M Plan: minimax.io/m-plan
MiniMax Code for desktop: agent.minimax.io/download
Zelig retweeted
MiniMax at SF Tech Week 🌉
📍 Oct 6 · Frontier Models, High-Speed Inference & Agents at Scale With @friendliai , @anacondainc , and @kilocode.
📍 Oct 7 · Open Source AI Stack With @novita_labs, @ArtificialAnlys, @NousResearch , @radixark , and @vercel.
📍 Oct 8 · Intelligence in the Open MiniMax research night with @FireworksAI_HQ , @radixark , @digitalocean , @Modular, and @evermind .
📍 Oct 8 · Magnific Tech: Growth at the Pace of AI MiniMax joins @magnific to discuss building and growing creative AI products.
See you in SF.
Details + RSVP 👇
Zelig retweeted
Open weights. Shared progress. MiniMax H3 is moving fast.
We built H3 for video generation with native stereo audio and multimodal reference control. The open-source community is making that capability faster, more accessible, and easier to build on.
Recent highlights:
• FastH3 — FastVideo, Nuva Lab and NVIDIA: 4-step distillation, now running on DGX Spark and Apple Silicon.
• Sol-H3 — NVIDIA’s SANA team: 15 seconds of 768p video + audio in 6.6 seconds on 8×B300, in the team’s warm-inference benchmark.*
• VDN — Haocheng Xi and the OpenVDN team: rethinking attention for faster H3 inference, with weights, training and inference code released.
• PDD — NVIDIA’s distillation method, brought to H3 by Alibaba PAI as 8-step Acc-LoRAs, now supported in ComfyUI.
• LightX2V — 4- and 8-step Turbo LoRAs, with workflows for text, image and reference-conditioned video + audio.
Behind every release are people training, optimizing, quantizing, testing and sharing. Special thanks to:
@haoailab @nuvalab @NVIDIAAI @xieenze_jr @HaochengXiUCB @ArashVahdat @julberner @LightX2V @ComfyUI
And to the individual contributors pushing the work forward:
@haozhangml @cxlcl1 @lawrence_cjs @yitongli165665 @haopengl33 @songhan_mit @shanasaimoe
Thank you for building with H3 and helping make it faster, more accessible, and more useful for the community. Powerful models go further when we build together. Keep pushing H3. Excited to see what comes next. 🚀
Explore the ecosystem: github.com/MiniMax-AI/awesom…
Zelig retweeted
@MiniMaxAgent just shipped H3-based plugins for dynamic images and white-model rendering! Been playing around with them a bit. We're still tuning the white model quality, but this is already pretty fun to mess with~
Zelig retweeted
🎵MiniMax-Music3
Next-Generation Open-Weights Production-Ready & Versatile Music Model
huggingface.co/MiniMaxAI/Min…
Zelig retweeted
Starting today, we’re opening the MMC TUI invite-only beta 🚀
Bring it into real repos, take on tough engineering tasks, and help us shape a coding agent built for the terminal. We’ll reset usage quotas for all beta participants later this week.
A huge thank you in advance to everyone joining us—we can’t wait to build this with you!
If u want to join our beta testing, please DM @MiniMaxAgent
Zelig retweeted
MiniMax Code v3.0.54 is live.
This one’s packed with some of our most-requested features, plus improvements across the app.
Here’s what’s new.
Zelig retweeted
Privacy WIN! #zkp #privacy #TornadoCash #zkvm
Summaries about the legal analysis, ca5.uscourts.gov/opinions/pu…
The International Emergency Economic Powers Act and the North Korea Sanctions and Policy Enhancement Act vest the President with the authority to regulate “property” in which a foreign “national” or “person” has an “interest.”
The court must determine if Tornado Cash and its smart contracts fall within this definition.
Analysis of “Property” Definition: The court starts with the statutory text and the ordinary meaning of “property,” which is something capable of being owned. The dictionary definitions and the Supreme Court's interpretation support this understanding. The immutable smart contracts at issue are not property because they are not capable of being owned. More than one thousand volunteers participated in a “trusted setup ceremony” to make the code unchangeable and uncontrollable, so no one can exclude others from using the contracts. Even with the sanctions, the contracts continue to operate, and some users may become liable when digital assets are transferred to them via Tornado Cash without their knowledge or consent.
OFAC's Regulatory Definition: The court also considers OFAC's regulatory definition of “property,” which includes “contracts of any nature” and “services of any nature.” However, the court finds that even under OFAC's definition, the immutable smart contracts do not qualify as property. They are not ownable, not contracts (as they lack an agreement between two or more parties), and not services (as they are merely tools used in providing a service and Tornado Cash does not own the services). The court rejects the Department's attempt to conflate “interest” with “property” and emphasizes that the catch-all for “any other property” still requires the item to be property.
Conclusion: The court holds that OFAC exceeded its statutory authority in designating Tornado Cash. The immutable smart contracts are not “property” under the common or regulatory definitions, so the court need not address whether Tornado Cash qualifies as an “entity” or has an “interest” in the contracts. The court reverses and remands the case to the district court with instructions to grant Van Loon's partial motion for summary judgment based on the Administrative Procedure Act.
Zelig retweeted
Thank you @vacp2p for the report - the more independent testing that takes place, the better. Here, we present some notes and clarifications:
Proof Size: zkMIPS uses Poseidon over Goldilocks, operating in a 64-bit field, compared to @SuccinctLabs or @RiscZero's 32-bit or 31-bit fields. While this increases proof size, it's not a major issue due to STARK-to-SNARK proof recursion, which compresses proofs to approximately 200 bytes for on-chain verification.
Memory Leakage: Our team are aware of the memory leakage issues reported, stemming from Plonky2's 'PolynomialBatch::from_values' requiring a value object instead of a trace-table reference, and we're currently working towards optimizing this in Plonky2.
Strategic Focus: ZKM's vision extends beyond Ethereum, and the zkMIPS SDK is being optimized for Bitcoin and multi-chain integrations. We aim to be the most production-ready, MIPS-based zkVM, making state-of-the-art ZK technology accessible to all.
We appreciate the comprehensive testing and look forward to ongoing improvements and collaboration in the zkVM space 🤝
Our latest blog explores our testing of 6 promising zkVMs to see which is most suitable for our Nescience zkVM architecture.
We assessed their memory consumption and ability to handle arithmetic operations.
Discover which we're most optimistic on: vac.dev/rlog/zkVM-testing
1/2
Zelig retweeted
🚀 AntChain OpenLabs #ZK #Workshop II
🔍 Explore the latest ZKP technology advancements and application challenges. Listen to global experts share insights, ZKP virtual machine acceleration experiences, and future trends.
📅 Sept. 24th, 9:30 AM (UTC+8), Online Live Stream
🔗 Register to Watch: n.dingtalk.com/dingding/live…
👥 Guest Details:
Ying Yan, Chief Scientist of Antchain OpenLabs
Srinath Setty @srinathtv, Principle Researcher of Microsoft Research
Sinka Gao @sinka2022, Founder of Delphinus Labs
Xiong (Leo) Fan @leofanxiong, Co-Founder of Cysic
Xiaohong Chen @_CHEN_Xiaohong_, CTO of Pi Square @pisquared
Stephen Duan @sd_eigen, Dev Lead of ZKM @ProjectZKM
Guiming Wu, Researcher of Antchain OpenLabs
Wishlonger @wishlonger, Co-founder of Pharos @pharos_network
Cobe Zhang, CEO of ZAN @zan_team
Ming Wu @spark_ren, CTO of 0G AI
Cecilia @ceciliaz030, Engineer of Taiko
Wisely, Principle Researcher of Antchain OpenLabs
📃 Agenda Details 👇
Zelig retweeted
Check the our latest post about Binius Stark Scheme. eigenlab.medium.com/binius-s… #zkp #stark #zkvm
