š WELCOME TO METAMESH.BIZ +++ OpenAI partnering with Cerebras to hit 750 tokens/sec because apparently GPT-5.6 wasn't fast enough, it needed to be fast enough at 14Ć +++ Google drops Gemini 3.7 Flash at $0.75/M input tokens while DeepSeek V4-Pro undercuts everyone at $0.435 because the race to zero is also accelerating +++ 30+ crypto firms complaining AI safety guardrails block their security teams while attackers roam free, which is the "locks only stop honest people" argument but with billion-dollar stakes +++ THE FUTURE COSTS LESS PER TOKEN EVERY WEEK AND NOBODY KNOWS WHAT TO DO WITH THE SAVINGS š ā¢
š WELCOME TO METAMESH.BIZ +++ OpenAI partnering with Cerebras to hit 750 tokens/sec because apparently GPT-5.6 wasn't fast enough, it needed to be fast enough at 14Ć +++ Google drops Gemini 3.7 Flash at $0.75/M input tokens while DeepSeek V4-Pro undercuts everyone at $0.435 because the race to zero is also accelerating +++ 30+ crypto firms complaining AI safety guardrails block their security teams while attackers roam free, which is the "locks only stop honest people" argument but with billion-dollar stakes +++ THE FUTURE COSTS LESS PER TOKEN EVERY WEEK AND NOBODY KNOWS WHAT TO DO WITH THE SAVINGS š ā¢
On August 13, 2026, Metamesh tracked 56 AI stories, including 3 clustered developments, and ranked them by signal rather than volume. The lead item was Researchers find that feeding a frontier model's encrypted reasoning traces to a weaker model from the same provider.... Also high in the stack: OpenAI previews Ultrafast, an API tier powered by Cerebras that runs GPT-5.6 Sol up to 14Ć faster and generates up... and Over 30 crypto companies, including Coinbase and Block, say frontier AI safety guardrails hinder legitimate security.... That combination is why this archive exists: it preserves the day's shape for AI practitioners, not just the last headline that crossed the wire.
The daily ticker's read: WELCOME TO METAMESH.BIZ +++ OpenAI partnering with Cerebras to hit 750 tokens/sec because apparently GPT-5.6 wasn't fast enough, it needed to be fast enough at 14Ć +++ Google drops Gemini 3.7 Flash at $0.75/M input tokens while DeepSeek V4-Pro undercuts.... Read against the ranked story list below, it gives the archive a point of view: what mattered, what was mostly noise, and which threads were worth saving for later comparison.
š You are visitor #47291 to this AWESOME site! š
Archive from: 2026-08-13 | Preserved for posterity ā”
š¬ "Any sort of evaluation with a sample size of 1 is essentially worthless"
ā¢ "Do not proxy results. Instead deploy the highest end model as a judge"
via Arxivš¤ Arda Uzunoglu, Benjamin van Durme, Daniel Khashabiš 2026-08-12
ā” Score: 8.1
"Large language models are increasingly trained and deployed with long contexts that span documents, code repositories, and interaction histories. This scaling reflects the implicit assumption that training on longer contexts will only help the model by exposing it to richer evidence. We challenge th..."
via Arxivš¤ Abigail Oppong, P Sam Sahil, Tadesse Destaw Belay et al.š 2026-08-11
ā” Score: 7.9
"Safety alignment in large language models (LLMs) is largely developed in English, assuming these safeguards generalize across multilingual settings. However, this assumption remains underexplored and exposes a vulnerability in low-resource languages. We investigate cross-lingual safety transfer in f..."
š¤ AI MODELS
DeepSeek V4-Pro Release
2x SOURCES šš 2026-08-12
ā” Score: 7.7
+++ DeepSeek's latest model trades blows with Kimi K3 on benchmarks while undercutting on cost by orders of magnitude, forcing an awkward reckoning about whether performance per dollar was always the real metric. +++
šÆ Benchmark vs. Reality ā¢ Cost Optimization ā¢ Model Reliability Concerns
š¬ "What benchmarks say, vs what I've been observing are different"
ā¢ "What I care about is whether the model is capable of the tasks I give it at the lowest cost"
š¤ AI MODELS
Google Gemini 3.7 Flash Launch
2x SOURCES šš 2026-08-13
ā” Score: 7.7
+++ Google launches yet another foundation model positioned as the efficient alternative, pricing it aggressively while the market figures out if "most intelligent workhorse" means anything when your last model already did the job. +++
š¬ "by Jan 2027 there will be at least 3 newer generation of models released already"
ā¢ "Gemini models...make mistakes, importantly, without correcting them for hallucinated API calls"
+++ An unreleased Claude variant made progress on Riemann hypothesis research while the team quietly improved safety filters on Fable 5, proving AI labs can multitask on capabilities and caution. +++
"An unreleased version of Claude has made strides on a problem related to the Riemann hypothesis. It improved the lower bound for the fraction of zeros of the Riemann zeta function that satisfy the hyp..."
šÆ Enterprise adoption controls ā¢ Educational AI benefits ā¢ Marketing vs substance
š¬ "ChatGPT has allowed her to augment her lesson plans with slides/games"
ā¢ "is this genuine progress or merely a marketing metric for stakeholders?"
šÆ Desktop app bloat ā¢ Performance degradation ā¢ Electron hypocrisy
š¬ "They've been updating UI to follow Claude patterns, and I've not been impressed"
ā¢ "A $1T company shouldn't make a shitty desktop app, especially one claiming to replace developers"
"Emergent misalignment (EM) is the phenomenon where fine-tuning a language model on a narrow task leads to harmful behavior in unrelated domains. A leading mechanistic account attributes EM to persona features: latent directions acquired during pre-training that misaligned fine-tuning amplifies. We a..."
via Arxivš¤ Yuzhong Shen, Masha Sosonkina, Peng Xu et al.š 2026-08-12
ā” Score: 6.8
"Modernizing legacy Fortran is a problem of volume: the transformations are individually routine, but the codebases can be enormous, and across much of computational science the work simply goes undone. We propose an agentic workflow that takes this work on at production scale, and we set out to meas..."
via Arxivš¤ Praveen Reddy, Charuta Mandke, Suvrankar Datta et al.š 2026-08-12
ā” Score: 6.8
"General-purpose large language models (LLMs) have recently been reported to match or exceed specialized clinical AI tools on medical benchmarks, but such comparisons draw on a narrow set of systems and on benchmarks developed largely in high-income settings. We evaluate VITA, a retrieval-augmented g..."
via Arxivš¤ Orr Paradise, Oliver Richardson, Yoshua Bengio et al.š 2026-08-11
ā” Score: 6.8
"When a probabilistic predictor answers many conditional-probability queries, are its answers self-consistent, and can this be verified in polynomial time? This problem is of interest for AI safety, where safety is derived from honesty about probabilistic predictions of unwanted outcomes potentially..."
via Arxivš¤ Sourabrata Mukherjee, Kalika Bali, Sunayana Sitaramš 2026-08-11
ā” Score: 6.7
"When a tool-using agent is given the same task in a different language, does it still take the same steps? Multilingual evaluation rarely asks: it compares final answers and discards the actions. Yet those actions are the product: they fix cost and latency, decide how the system fails, and are the o..."
"About a year ago, David and I put up two bounty problems involving natural latents. I am now about 80% confident that both have been resolved, both wā¦..."
via Arxivš¤ Alan Li, Rahul Saha, Anton Xue et al.š 2026-08-11
ā” Score: 6.7
"AI agents are increasingly used in mathematics research, but it is often unclear how to use them effectively. Towards this, we present an extensive case study of how AI was used to improve bounds on the Grothendieck constant $K_G$, which captures the hardness between combinatorial problems and their..."
via Arxivš¤ Jean-Pierre Busch, Guido Linden, Jan Bergmann et al.š 2026-08-12
ā” Score: 6.7
"Recent research in machine and deep learning has shown the potential of learningbased motion planning approaches to improve the driving behavior of automated vehicles, especially in complex environments. However, their complex nature and lack of transparency can hinder explainability and trustworthi..."
via Arxivš¤ Simon Yu, Nicholas Tomlin, Marwa Abdulhai et al.š 2026-08-12
ā” Score: 6.6
"Multi-agent reinforcement learning for human-AI interaction typically relies on a single large language model to simulate user behavior. We show that this approach systematically fails to generalize, and trace the failure to simulator collapse: because the simulator LLM is mode-collapsed, an LLM pol..."
via Arxivš¤ Aman Tyagi, Hemanth Boinpally, Jonathan Chen et al.š 2026-08-12
ā” Score: 6.6
"Modern black-box Image-to-Video (I2V) models offer powerful capabilities in automated content creation, yet their lack of fine-grained control and reliability presents significant challenges in professional workflows. Their inherent stochasticity causes minor variations in textual prompts or hyperpa..."
via Arxivš¤ Ankita Rajaram Naik, Anupama Murthi, Benjamin Elder et al.š 2026-08-12
ā” Score: 6.6
"Agents deployed in enterprise settings must reason across structured APIs and document collections, yet existing benchmarks evaluate these capabilities in isolation. We introduce VAKRA (e\textbf{V}aluating \textbf{A}PI and \textbf{K}nowledge \textbf{R}etrieval \textbf{A}gents), a benchmark of over $..."
via Arxivš¤ Aleksandra Kalisz, Jack Simons, Krisztina Sinkovics et al.š 2026-08-12
ā” Score: 6.6
"Foundation models for protein structure prediction remain unreliable on certain targets. External oracles can flag and correct these failures, but biological oracles are expensive, making oracle budget a critical constraint. Existing guidance methods, such as FK-steering, DPO, and Best K-of-N sampli..."
"Agentic coding READMEs like CLAUDE.md grow without bound in real repositories, stopping only when the repository retires or someone rewrites the file wholesale. We trace this to imperfect recall: appending an instruction is always cheap, but once an instruction's rationale is gone, deleting it witho..."
via Arxivš¤ Cheng Qian, Wenting Zhao, Liangwei Yang et al.š 2026-08-12
ā” Score: 6.5
"Recent work on distillation transfers the capabilities of large models to smaller ones often by updating the latter's parameters, through teacher forcing, on-policy distillation, and related training-time methods. In this paper, we ask whether such transfer can instead occur at test time. We study s..."
via Arxivš¤ Antoine de Mathelin, Christopher Tosh, Wesley Tanseyš 2026-08-12
ā” Score: 6.5
"Treating patients with combinations of drugs reduces the risk of resistance to any individual drug. Finding effective combinations is difficult because the large search space makes combinatorial screens prohibitively expensive, time consuming, and often technically infeasible. Predictive models can..."
via Arxivš¤ Yuchao Wu, Junqin Li, XingCheng Liang et al.š 2026-08-12
ā” Score: 6.5
"While retrieval-augmented generation (RAG) has proven effective at giving LLMs access to external knowledge, mainstream dense-retrieval implementations remain inherently limited in handling structured constraints and multi-hop reasoning. Graph-based methods address this by constructing knowledge gra..."
via Arxivš¤ Zetao Hong, Song Yuan, Yuanhao Ding et al.š 2026-08-11
ā” Score: 6.5
"Modern reinforcement learning (RL) post-training pipelines for large language models (LLMs) increasingly combine rollout workloads across multiple domains and feedback paradigms. Prefix-aware routing improves inference efficiency through cache reuse and load balancing, but it does not control how he..."
via Arxivš¤ Dong Qiao, Chris Ding, Jicong Fanš 2026-08-11
ā” Score: 6.5
"Benchmark leaderboards summarize how well a language model performs, but not how its behavior relates to that of other models or changes across generations. We characterize the output behavior of 32 models from six families using their responses to a shared bank of 10{,}000 prompts. After embedding..."
via Arxivš¤ Yilin Liu, Rui Meng, Wangze Ni et al.š 2026-08-12
ā” Score: 6.4
"Retrieval-Augmented Generation (RAG) repeatedly prefills identical text chunks across queries, incurring redundant computations. Position-Independent Caching (PIC) mitigates it by reusing precomputed Key-Value (KV) across positions, but its efficiency is constrained by the large volume of text token..."
"Like many others, I felt surprised and alarmed by the recent wave of revelations about LLM agents hacking real systems during training episodes and eā¦..."
š¬ "Systems of authority tend to react poorly to pentests, whether authorized or unauthorized"
ā¢ "retconning the satisfaction of one's desired outcome as altruistic intent"
via Arxivš¤ Minsoo Kim, Sungyoung Ji, Kisung Moon et al.š 2026-08-11
ā” Score: 6.1
"We propose that a model's uncertainty about a token is reflected not only in the breadth of its output distribution but also in whether a confident prediction is \emph{fragile} under perturbation of its attention pathways. We instantiate this as ASMI (Attention-Subnetwork Mutual Information), a trai..."
via Arxivš¤ Di Yang Shi, W. Bradley Knoxš 2026-08-12
ā” Score: 6.1
"We present a formal process to enable non-experts to instantiate and iterate on human-aligned reward functions, i.e. reward functions that adhere to a given preference ordering over trajectories. Given a task described in natural language, our process produces a linear reward function in three steps..."
via Arxivš¤ Weihao Bo, Shan Zhang, Yanpeng Sun et al.š 2026-08-12
ā” Score: 6.1
"Multimodal Large Language Models (MLLMs) have been growing the capability for scientific writing and collaboration. For example, OpenAI Prism is a free workspace for scientific writing and collaboration. One important feature in Prism is turning scientific diagrams directly into LaTeX TikZ code. In..."