ZeroHour
DashboardNewsCVEsExploitsIncidentsAIBriefAnalyticsEntitiesWatchlistIOCsAsk the feedSourcesFeeds & APIKiosk mode

Search: “tokenization”

484 stories in the last 30d

Searchkeywordsmart Sortlatesthotimportancerelevance
DomainAllCyberAIWindow24h3d7d30dAllIn the wildImportant onlyShow duplicates
CategoryAllAI research 85AI safety & security 70Exploit / PoC 56Model release 50Malware 43Threat actor 43AI industry 42Vulnerability 30Phishing & fraud 28Research 26AI tools & infra 20Advisory 15Industry 14Data breach 14
SeverityAllcritical 40high 117medium 94low 2info 304

Speculative Decoding in vLLM on AMD GPUs

vLLM benchmarks speculative decoding on AMD Instinct MI300X and MI355X GPUs across five drafting methods including EAGLE-3 and native MTP.

The vLLM project documents draft-and-verify speculative decoding support for AMD GPUs via ROCm, comparing native MTP, Gemma 4 MTP, EAGLE-3, DFlash, and DSpark drafting approaches. Output-token throughput effects varied with drafting method, proposal length, model family, draft checkpoint, workload, and acceptance behavior. The post also covers how to enable each method plus practical tuning and observability considerations.

Hacker News · AI · 8d agoAI tools & infraHN 141↑ · 52 comments1

Rare Not Random Using Token Efficiency for Secrets Scanning

Researcher proposes token efficiency (string length divided by BPE token count) as a better post-regex filter than entropy for secrets scanning, validated on CredData.

The post explores whether Byte-Pair Encoding tokenization can replace Shannon entropy as the primary filter for candidate secrets captured by regex in tools like Gitleaks. It defines 'token efficiency' as string length divided by token count under the cl100k_base tokenizer; secret-like strings such as GitHub tokens tokenize into many small tokens and score low, while natural text scores high. Evaluating labeled secrets from the CredData dataset shows a usable separation, with roughly 2.5 suggested as a minimum cutoff versus Gitleaks' 3.5 entropy threshold. The technique is positioned as a post-regex filtering step rather than a standalone detector.

Lobsters · security · 4d agoResearch

We got admin access to Baseten's production GitHub in 25 minutes

Strix autonomous hacking agent extracted a working GitHub token with repo admin rights from Baseten's public Harbor image; Baseten rotated it next day.

Strix, an autonomous hacking agent, scanned *.baseten.co without credentials and found a public Harbor container registry project anonymously exposing the baseten/baseten-app image. A GitHub personal access token for basetenbot, embedded in Docker build history since March 2023, still worked in July 2026 and granted admin/push rights to basetenlabs/baseten, flux-cd, and homebrew-tap plus read/write on private customer repos. Baseten, valued at $13 billion, confirmed the issue as critical and rotated the token within a day.

Hacker News · AI · 14h agoVulnerabilityHN 51↑ · 12 comments1

NIST and CISA finalize playbook to stop token theft and forgery

NIST and CISA finalized NIST IR 8587, a playbook helping federal agencies and cloud providers defend identity tokens against theft and forgery.

The finalized NIST IR 8587 guidance covers protecting token signing keys, verifying tokens, lifetimes, revocation, session management, and dividing security responsibilities between cloud providers and customers. It cites an incident in which foreign actors forged tokens with a stolen commercial signing key to steal more than 60,000 emails from one government agency. It also recommends extending token protections to AI agents and preparing identity systems for a future post-quantum cryptography transition.

Help Net Security · 1h agoAdvisory 2 sources

A Princeton Researcher Proposes Recurrent Looped Transformer (RLT) that Carries Decoder State across Every Token, Fixing 96 Blocks per Token with Unbounded Temporal Depth

Princeton researcher Yifan Zhang proposes Recurrent Looped Transformer, carrying full decoder state across every token for unbounded temporal depth.

Yifan Zhang's technical report defines the Recurrent Looped Transformer (RLT), pairing a causal encoder with a recurrent decoder whose final output and layerwise sliding-window attention cache carry into every subsequent token with no prompt-response boundary reset. The reference configuration ties 48 encoder and 48 decoder layers, executing 96 logical blocks per token while the state path grows to 48t blocks after t tokens at fixed per-token compute. The report details RL replay contracts that rebuild all states under current parameters and exact prefix snapshots for multi-turn serving, but explicitly reports no measured efficiency, reasoning quality, or scaling results.

MarkTechPost · 2d agoAI research1

Malicious Twitch Extension Exposes OAuth Tokens of 30,000 Chrome and Firefox Users

Malicious 'Twitch Enhanced Viewer | JeetBot' browser extension stole live OAuth session tokens from roughly 31,000 Chrome and Firefox users.

Socket.dev found the cross-store extension 'Twitch Enhanced Viewer | JeetBot' rerouted Twitch playlist requests through operator-controlled proxies, attaching the victim's account-level OAuth token as a URL parameter where it could be logged in cleartext. About 30,000 Chrome and 552 Firefox installs were exposed, with tokens forwarded for nearly every watched channel to infrastructure tied to a Russian commercial bot service. The stolen bearer token enables chat, whisper, account-setting access and channel-point spending without the password or 2FA.

Cyber Security Newsupdated · 1d agofirst · 1d agoMalware in the wild 2 sources

Malicious Twitch Browser Extension Leaks OAuth Tokens From Nearly 31,000 Users

Malicious Twitch extension 'Twitch Enhanced Viewer | JeetBot' leaked OAuth tokens from nearly 31,000 users to Russian bot-service proxy servers.

Socket researchers found the 'Twitch Enhanced Viewer | JeetBot' browser extension forwarded users' live Twitch OAuth tokens as an &auth= query parameter to operator-controlled proxy servers, affecting about 30,000 Chrome users and 604 Firefox users. The tokens, written in cleartext to proxy logs, are bearer credentials granting access to chat, whispers, and account settings without password or 2FA. The operator is a Russian commercial Twitch/Kick/VK-Live bot SaaS linked to Cyprus-based developer Aleksandr Popov; earlier v4.x builds also POSTed tokens to a set-token endpoint. Firefox add-on version 85.8.7 stops the token forwarding, but previously transmitted tokens remain valid until revoked.

The Hacker News · 2d agoMalware in the wild

[AINews] Claude Fable/Mythos 5.1: new SOTA model, 75% cache price cut but 70% more output tokens

Anthropic launched Claude Fable 5.1 and Mythos 5.1, claiming new SOTA benchmarks, with 75% cache-read price cut and 1M-token context.

Anthropic released Claude Fable 5.1 and Mythos 5.1 as flagship models for coding and knowledge work, with a 1M-token context window and pricing of $10/$50 per million input/output tokens and cache reads cut 75% to $0.25. Artificial Analysis Intelligence Index scored Fable 5.1 at 66 versus 63 for Claude Opus 5, with HLE at 59.1% and Terminal-Bench v2.1 at 91.4%, though per-task cost rose ~20% due to 1.7x output token usage. Community analysis suggested Fable and Mythos may share underlying weights with different safety/routing behavior, and release notes highlighted Enterprise Frontier Safeguards and zero-data-retention support.

Latent Space · 14d agoModel release2

CISA and NIST Release Technical Checklist for Safeguarding Identity Tokens From Theft and Misuse

CISA and NIST published NIST IR 8587, final guidance for protecting identity tokens from forgery, theft, replay, and signing-key compromise.

NIST Interagency Report 8587 (September 15, 2026) expands the IA-13 'Identity Providers and Authorization Servers' control from NIST SP 800-53 R5.1.1, guiding federal agencies and cloud providers on SSO, identity federation, and machine-to-machine authentication. It requires hardware-backed signing-key storage for moderate-impact systems, 90-day key rotation for high-impact systems, token lifetimes under one hour, and sender-constrained mechanisms such as mutual TLS and DPoP. The report cites incidents including forged SAML assertions that exposed over 60,000 emails from a federal agency. It also extends guidance to agentic AI systems using signed tokens and urges post-quantum cryptography migration planning.

Cyber Security News · 15h agoAdvisory1

Inside Knight Office, a New M365 AiTM Phishing Kit

Huntress details Knight Office, a new AiTM phishing kit stealing Microsoft 365 session tokens and registering attacker-controlled devices.

Huntress investigated an August 18, 2026 adversary-in-the-middle attack using the Knight Office phishing kit, which captured Microsoft 365 session tokens via a DocuSign-style lure routed through a Monday.com tracking link and a compromised Joomla site. After token capture, the actor enrolled an unauthorized host in Microsoft Entra ID and bound a Windows Hello for Business key credential to the account. The kit's operator console, found at IP 104.37.188.94, is built on Python Flask and hosts at least 25 .vu phishing domains. Telemetry links the kit to at least nine token-replay logins in two weeks and roughly 700 reported lure emails since April.

Huntress · 13d agoPhishing & fraud in the wild1

Training a 3.8B LLM to 0.384 CORE for $998 – Hugo Vergnes

Independent developer Hugo Vergnes trained a 3.8B-parameter Llama-style model to 0.384 CORE on 65B tokens for $998 in 43 hours on rented B200s.

Hugo Vergnes trained little-lm, a 3.848B-parameter decoder-only LLM, on 65.3B tokens in 43 hours for $998 using rented NVIDIA B200s, scoring 0.384 on the CORE benchmark and beating nanochat d32 (0.310) at similar cost. The Llama-style architecture uses RMSNorm, RoPE, GQA with 24 query and 8 KV heads, relu-squared MLPs, QK-norm, and ResFormer-style value embeddings that account for 19% of parameters. Key wins included the Muon optimizer for matrix parameters, a trapezoidal learning-rate schedule with linear cooldown, FP8 training plus vocabulary padding for roughly 33% throughput gains, and the ClimMix dataset over FineWeb-Edu. The project, inspired by Karpathy's nanochat, was built as a config-driven YAML framework for small LLM training.

Hacker News · security · 6d agoAI researchHN 28↑ · 2 comments1

AI Infra Summit: NVIDIA Vera Rubin and DSX Platform Advancements Showcase Energy Efficiencies of Optimizing Tokens Per Watt for AI Factories

At AI Infra Summit, NVIDIA showcased Vera Rubin and DSX gains up to 1.4x tokens per megawatt, plus Annapurna, d-Matrix, and Pinterest partnerships.

Ian Buck's AI Infra Summit keynote before 8,000+ attendees emphasized validated agentic tokens per megawatt as the emerging AI infrastructure metric. Announcements include Amazon's Annapurna Labs collaborating on NVHBM custom high-bandwidth memory, d-Matrix integrating NVLink Fusion with Raptor XPUs, and Pinterest using Blackwell plus Dynamo inference software for conversational visual discovery. Lambda reported 23% better performance per watt with DSX MaxLPS on Blackwell servers, running 19 nodes on a 16-node power budget. NVIDIA says DSX MaxLPS combined with Groq 3 LPX on Vera Rubin NVL72 targets up to 35X token throughput per megawatt versus GB200 NVL72 for 2-trillion-plus-parameter models.

NVIDIA Blog · 16h agoAI industry

Infostealer Logs Expose Replayable AI Tokens That Can Bypass MFA

Okta finds infostealer logs contain thousands of replayable AI session tokens and API keys, letting criminals bypass MFA and access services from Google, Anthropic and OpenAI.

Okta analyzed a 7 GB infostealer dump from August 2, 2026 covering 5,871 infected machines in 162 countries and found 555 of 44,791 JWTs related to AI services, plus 1,843 unexpired JWTs and JWEs (largely set by OpenAI via NextAuth.js) and 24 still-valid API keys for Google Gemini, OpenAI, Groq and OpenRouter. Valid session tokens and API keys can be replayed with anti-detect browsers like Camoufox to bypass credential and MFA checks, fueling an underground market for AI account access known as LLMjacking, where attackers rack up victims' AI compute bills. Some 17.7% of the JWTs contained plaintext PII usable for social engineering. Google's GTIG reported growing buyer demand for Claude, Gemini, Cursor and Devin credentials, and Mandiant handled an incident where an actor used an exposed GitHub PAT to deploy unauthorized AI infrastructure and scale high-performance compute.

The Hacker News · 6d agoThreat actor in the wild 2 sources1

Ask HN: How do you manage skills files?

A Hacker News thread debates whether agent skill files are worth managing, citing 2–4x output-token reductions on flagship models in one company's testing.

Commenters argue skills are stored prompts that help less-technical users compensate for weak prompting, while one participant reports company testing found skills reduce flagship-model output tokens by roughly 2–4x, a gap growing with newer models. Others note skills can bundle reusable scripts and inline commands for deterministic context building, and that harnesses now execute backticked commands before the agent sees the skill. Some argue improving model capability makes downloadable skills redundant.

Hacker News · security · 9d agoAI tools & infraHN 303↑ · 271 comments

ukisai/Swift-Qwen3.8-27b — new model trending #30 on Hugging Face

UkisAI releases Swift-Qwen3.8-27B, a Qwen3.8-27B derivative using 58.3% fewer thinking tokens with <1% performance loss and ~1.95x speed-up.

UkisAI released Swift-Qwen3.8-27B, a reasoning-efficient derivative of Qwen3.8-27B that cuts thinking-token usage by 58.3% while staying within 1% of base performance, yielding a 1.95x speed-up on several tasks. The model was fine-tuned by penalizing reasoning-marker tokens that trigger overthinking, plus a transfer component from BottleCap AI's ThinkingCap-Qwen3.6-27B. Benchmarks include GPQA-Diamond 88.28% (base 88.38%), MMLU-Pro 84.95% (base 85.47%), and AIME 2026 94.00% (base 98.67%), with mean-token reductions of roughly 27-46% across tests. GGUF weights are available on Hugging Face alongside enterprise licensing options.

Hugging Face trending models · 7d agoModel release

deepseek-ai/DeepSeek-V4.1-Flash — new model trending #28 on Hugging Face

DeepSeek releases DeepSeek-V4.1-Flash, a 552B-parameter multimodal MoE model with 1M-token context and KV cache cut to 890 bytes per token.

DeepSeek-V4.1-Flash is a multimodal Mixture-of-Experts model with a 552B-parameter backbone that activates 8B parameters per token during prefill and 16B during decode. It uses a Causal Encoder-Decoder architecture, Compressed Sparse Attention 2, and FP4 KV caching to reduce the global KV cache footprint to 890 bytes per token, roughly one quarter of DeepSeek-V4-Flash. The model was trained from scratch on 45T tokens with context extended to 1M tokens, includes an Engram conditional-memory module (196B parameters), and is released under the MIT license. Post-training uses SFT, RL, and on-policy distillation with large-scale automated synthesis of agentic tasks and a controllable reasoning effort setting from 1 to 100.

Hugging Face trending modelsupdated · 5d agofirst · 6d agoModel release 3 sources1

Studying Image Tokenizers as Visual Languages in Unified Multimodal Models

A controlled pure-autoregressive testbed shows task-specific validation losses rank image tokenizers differently, with I2T loss the most consistent signal.

Researchers built a controlled pure-autoregressive testbed and tracked task-specific validation losses during multimodal continual pretraining across text, image, text-to-image (T2I), and image-to-text (I2T) prediction. They find losses should be analyzed per task because they exhibit distinct scaling behavior and rank tokenizers differently, and that the loss-performance relationship depends on the predicted token space. I2T loss, computed over a shared text vocabulary, correlates consistently with both generation and visual understanding performance after supervised finetuning. Case studies revisit the discriminator, semantic supervision, and vocabulary size as tokenizer design axes.

arXiv cs.AI / cs.LG / cs.CL · 7d agoAI research1

BigBear 2.0 Evilginx2 Phishing Campaign Bypasses Microsoft 365 MFA With Session Cookie Theft

CloudSEK identified BigBear 2.0, an Evilginx2-based AiTM phishing operation stealing Microsoft 365 MFA session cookies, hitting 461 organizations across 40-plus countries.

The campaign proxies Microsoft sign-in pages to capture credentials and authenticated session cookies, enabling session replay into email, Teams, SharePoint, OneDrive, and connected SSO applications. CloudSEK's June 2026 discovery found 5,137 stolen records, 1,032 passwords, and 4,148 session cookies tied to 3,331 victim IPs, linked to operator 'General Boss' across 42 VPS nodes. The operation targeted IT services and managed service providers, used country-matched residential proxies, and involved at least five affiliates.

Cyber Security News · 8d agoPhishing & fraud in the wild1

Studying Image Tokenizers as Visual Languages in Unified Multimodal Models

A controlled autoregressive testbed shows validation losses must be analyzed per task, and image tokenizer choice affects joint multimodal text modeling.

Researchers built a pure-autoregressive testbed to study image tokenizers as the 'visual language' of unified multimodal models, tracking task-specific validation losses during multimodal continual pretraining across text, image, text-to-image (T2I), and image-to-text (I2T) prediction. They found that losses exhibit distinct scaling behavior per task and rank tokenizers differently, and that I2T loss over a shared text vocabulary gives a more consistent loss–performance signal than T2I loss. Better reconstruction does not necessarily yield lower task-specific losses or stronger downstream performance, and tokenizer choice can affect text modeling under joint optimization. Case studies examine the discriminator, semantic supervision, and vocabulary size design axes.

Hugging Face daily papers · 8d agoAI research1

CoVeR: Coverage-Based Token Pruning for Multi-View 3D Reasoning in VLMs

CoVeR, a training-free coverage-based token pruner, preserves 93.5% of VLM 3D-reasoning performance using only about 8% of visual tokens.

Researchers introduce CoVeR, a deterministic, training-free selector that chooses visual tokens to cover every region of a multi-view 3D scene using only token coordinates. Unlike learned-importance and voxelization pruners, it enforces an exact per-scene token budget, avoids saturation plateaus, and prevents near-duplicate selections. Experiments across four vision-language models show it surpasses prior state of the art by 3.9 percentage points on average across three 3D reasoning benchmarks.

Hugging Face daily papers · 8d agoAI research

Attackers Chain JFrog Artifactory Flaws to Gain Admin Control and Plant Backdoors

Attackers chained JFrog Artifactory flaws CVE-2026-42018 and CVE-2026-42016 for admin control, planting Rust backdoors; CVE-2026-82329 also mass-exploited.

Wiz observed attackers chaining CVE-2026-42018 (anonymous-token leak) and CVE-2026-42016 (token escalation) in self-hosted JFrog Artifactory between August 15 and September 8, gaining admin control, creating admin accounts, and installing malicious Groovy plugins and a custom Rust backdoor. Separately, critical authentication bypass CVE-2026-82329 (CVSS 9.8) was mass-exploited starting September 1, drawing ~406,000 exploitation attempts on September 2 per Fastly. CISA added CVE-2026-82329 to KEV on September 2 with a September 5 federal deadline. Patching does not revoke minted tokens or rotate stolen join keys; admins must review accounts and rotate credentials.

The Hacker Newsupdated · 1d agofirst · 5d agoExploit / PoC in the wild 7 sourcesCVE-2026-42018CVE-2026-42016CVE-2026-823292

Agnes-AI/Agnes-3.0-Flash — new model trending #30 on Hugging Face

Agnes AI releases open-weight Agnes-3.0-Flash Preview, a 33B multimodal model with 262k-token context under Apache 2.0.

Agnes AI released Agnes-3.0-Flash Preview, an open-weights multimodal checkpoint with 33B parameters and a 262,144-token context window under Apache 2.0. The model supports text, image, and video understanding, tool calling, and adjustable reasoning effort. The repo clarifies this preview checkpoint is distinct from the production/API Agnes 3.0 Flash model, which uses a different configuration with a 1M-token context window. Reported reference results include IFBench 74.20 and SciCode 38.08 against peers such as Qwen3.6-35B-A3B, Kimi K2.5, and MiniMax M3.

Hugging Face trending models · 4d agoModel release

Hackers are stealing Claude tokens from subscribers

Infostealer malware is stealing Claude login sessions, letting attackers mint OAuth tokens and burn subscribers' paid usage largely undetected.

Anthropic confirmed a bad actor used common infostealer malware to steal Claude login sessions from users' computers and consume their paid usage. A UK consultant saw idle token usage climb, and Anthropic suspended his account, invalidated sessions and Claude Code tokens, and issued a £44.49 partial refund on his $200-per-month plan. Multiple other users on Reddit and GitHub reported similar theft; Anthropic signed out affected users and issued refunds, but still lacks itemized usage reporting to help users detect misuse.

TechCrunch · AI · 7d agoAI safety & security in the wild1

Show HN: LLM Attention Visualization

A developer released a browser-based tool that visualizes which past tokens influence each LLM output token using aggregated, value-weighted attention scores.

A Show HN project presents a React application built on Transformers.js that renders per-token attention influence by aggregating attention weights scaled by value-vector magnitudes across all attention heads and layers. To expose internal tensors, the author instrumented the ONNX computation graph, hosted a modified model on Hugging Face, and pre-generated prompts to avoid long model downloads in the browser. Demos with a 600-million-parameter model show how verbatim copying draws heavily on source tokens and how single outputs blend information from multiple phrases.

Hacker News · AI · 7d agoAI tools & infraHN 41↑ · 9 comments

DavidAU/Qwen3.8-27B-TURBO-Fable-Cold-Fusion-735-882-Heretic-Uncensored-NEO-CODER-MAX-MTP-GGUF — new model trending #8 on Hugging Face

A new Qwen3.8-27B GGUF fine-tune claims ARC-C 735 at 8-bit with thinking tokens cut 2x-10x versus the base model.

Independent creator DavidAU released a GGUF fine-tune of Qwen3.8-27B built with Unsloth, claiming ARC-C of 735 at 8-bit and 719 at 4-bit, trending #8 on Hugging Face. The 'TURBO' variant cuts thinking tokens by one half to as much as one tenth while retaining output quality and detail. The repo ships both regular and MTP quants and claims gains over the base model across seven benchmarks, using 'Cold Fusion (GAIN + Unsloth)' and 'Fable Fusion 711' training methods.

Hugging Face trending models · 15d agoModel release

Twitch extension with 30K installs exposes users’ OAuth tokens

Twitch extension with 30,000+ installs exfiltrates users' OAuth session tokens to Russian-run JeetBot proxy servers.

Socket analysis shows the "Twitch Enhanced Viewer | JeetBot" browser extension, with over 30,000 installs on the official Chrome and Firefox stores, captures the Twitch web client's authorization header and extracts the user's OAuth token. The token is appended as an auth= URL parameter to video playlist requests routed through JeetBot proxy servers, landing in cleartext request logs retrievable by the Russian-language bot service vendor; ten hardcoded Russian-language channels are exempted. Earlier versions used more explicit token exfiltration, and the extension remained live in both stores at publication. Socket recommends removing the extension, disconnecting all Twitch sessions, and re-authenticating.

BleepingComputer · 1d agoMalware in the wild

Don't Send What You Don't Need: Question-Guided Token Pruning as a Privacy Defense for Vision-Language Models

QPriv-VL prunes privacy-sensitive visual tokens in federated/split VQA, cutting membership-inference success on VQA-RAD from 0.99 to 0.76-0.79 using ~40% of tokens.

The paper proposes QPriv-VL, a question-guided token-pruning framework for federated, split, and U-shaped split learning that suppresses privacy-sensitive visual patches before transmission. Its Dynamic Threshold Predictor combines cross-modal question relevance with frozen DINOv2-derived sensitivity to compute a per-sample pruning ratio and retention mask in one forward pass, without sensitivity labels. Evaluated on GQA, OK-VQA, VQAv2, SLAKE, VQA-RAD, and PathVQA against FSHA, FORA, iDLG, and attribute-inference membership inference attacks, it matches or beats fixed-ratio pruning. On VQA-RAD it reduces membership-inference success from 0.99 to 0.76-0.79 while preserving competitive accuracy with about 40% of the original token budget.

arXiv cs.CR · 1d agoResearch

Jackrong/Qwopus3.8-27B-Flash-GGUF — new model trending #26 on Hugging Face

Community fine-tune Qwopus3.8-27B-Flash, built on Qwen3.8-27B, cuts agent reasoning latency with 12.8% faster decoding and 80.7% MTP acceptance.

Jackrong released Qwopus3.8-27B-Flash, a fine-tune of Qwen3.8-27B optimized for long-running agent workloads, reporting 12.8% faster decoding and 80.7% multi-token-prediction acceptance. Training used roughly 1.5 million teacher-scored SFT examples filtered to the top 10%, followed by reinforcement training with NVIDIA NeMo-RL and GSPO. The author notes an explicit trade-off: MMLU-Pro mixed-set scores are lower than the base model, and a known bug can produce incorrect Python indentation. Author-provided benchmarks have not been independently verified.

Hugging Face trending models · 11d agoModel release1

Why don't machine learning research agents overfit?

Amazon researchers explain why ML research agents avoid benchmark overfitting, attributing generalization to compressibility of successful strategies.

Amazon Science summarizes the paper "What fits (into few tokens) doesn't overfit: Compression and generalization in ML research agents," which investigates why benchmark hill-climbing loops, whether run by human communities or LLM research agents, do not produce rampant overfitting. The explanation formalizes Occam's razor via a counting argument: successful ML strategies are highly compressible, so short descriptions lack room to memorize benchmark data and must capture real structure. LLM-based agents, being resettable and controllable, allow this hypothesis to be tested empirically.

Hacker News · security · 1d agoAI researchHN 21↑ · 3 comments

Context Engineering Inside the Harness: 4 Mechanisms That Beat Context Overflow and Goal Loss on Long-Horizon Tasks

Survey of four harness mechanisms—context budgeting, compaction, todo-state, and memory—that keep long-horizon LLM agents on task across 200+ tool calls.

The article details how agent harnesses, not larger context windows, solve context overflow and goal loss on long-horizon tasks, citing Chroma's Context Rot report showing 18 LLMs (GPT-4.1, Claude 4, Gemini 2.5, Qwen3) degrade on long inputs. Concrete implementations include LangChain Deep Agents offloading tool responses over 20,000 tokens to the filesystem and truncating old tool calls at 85% window usage, and Claude Code capping auto memory at 25KB while re-reading the 5 most recently modified files after compaction. OpenAI's Responses API now offers server-side compaction via context_management with a standalone /responses/compact endpoint, which Codex uses for long-running coding tasks. Manus reports a roughly 100:1 input-to-output token ratio per ~50-tool-call task, motivating todo.md state recitation to prevent goal drift.

MarkTechPost · 3d agoAI research1

[AINews] DeepSeek v4.1-Flash: 763B-P8B-D16B novel causal Encoder–Decoder architecture with vision marks the Return of the Whale

DeepSeek released V4.1-Flash, an open-weight 763B-parameter model with a novel causal encoder-decoder architecture, 1M context, vision input, and MIT license.

DeepSeek launched V4.1-Flash, an open-weight MIT-licensed model using a novel causal encoder-decoder architecture with 763B total parameters and asymmetric active parameters: 8B for prefill and 16B for decode. It supports 1M-token context and text+image input, priced at $0.30 per 1M input and $1.20 per 1M output tokens with a 50% off-peak discount. Artificial Analysis scored it 40 on its Intelligence Index, above DeepSeek V4 Pro 0813, and Vals ranked it the #1 open-weight model ahead of Kimi K3. Baseten shipped day-0 support and Ollama began rolling it out to paid subscribers.

Latent Space · 4d agoModel release1

Meet Redis LangCache: A Managed Semantic Cache That Cuts LLM API Costs by Up to 90% and Returns Cache Hits Up to 15x Faster

Redis launches LangCache, a managed semantic cache matching LLM prompts by meaning, cutting API costs up to 90% and returning hits up to 15x faster.

Redis LangCache is a fully managed semantic caching service in public preview on Redis Cloud, accessed via a REST API with Python and JavaScript SDKs. It embeds incoming prompts, vector-searches stored entries, and returns a cached response when similarity clears a configured threshold, skipping the LLM call entirely. Redis claims up to 90% cost savings and up to 15x faster cache hits; a demo run showed 0.37 seconds versus 2.232 seconds direct inference (about 6x) with zero LLM tokens. Customer Mangoes.ai reports a 70% hit rate, 70% lower LLM spend, and 4x faster responses on a patient-care voice app.

MarkTechPost · 5d agoAI tools & infra

[AINews] Andrew Ng gets into AI Engineering

Andrew Ng relaunches DeepLearning.AI around AI Engineering, defining four core skills from an analysis of 10,000+ job postings and expert interviews.

Andrew Ng, cofounder of Google Brain and Coursera, relaunched DeepLearning.AI with a focus on AI Engineering, basing the curriculum direction on an analysis of over 10,000 job postings plus interviews and surveys. He identifies four key skills: building and deploying AI applications, software engineering fundamentals, effective use of coding agents, and shaping the build with product sense. The Latent Space AI News issue also recaps agent ecosystem developments, including NVIDIA's 'Skill Lift' evaluation proposal showing skill scan scores correlate only weakly (Spearman rho = 0.14) with judged quality, and Konwinski's open-source persistent-agent 'microharness' Headlong, which achieved an unattended self-debugging repair in 48 minutes.

Latent Space · 22d agoAI industry1

Give every teammate and agent the right level of access to your Workers

Cloudflare launches per-Worker granular access controls with four roles, enabling least-privilege access for teammates, AI agents, and CI/CD pipelines.

Cloudflare announced granular authorization for Workers, letting admins scope access to a single Worker instead of the whole account. Four new roles are available: Metadata Read-Only (observability without source code), Content Read-Only (read code without changes), Editor (deploy without delete), and Admin (full control of one Worker). Roles apply at Developer Platform, product, or resource level, can be attached to dashboard users or API tokens, and are available to all customers now, with plans to extend to D1, R2, and KV.

Cloudflare Blog · 19h agoTools

GPT-6 Astra, Looped Transformers, and Hidden Reasoning

OpenAI released GPT-6 Astra, its strongest model to date, with standout 3D rendering and computer-use performance and 99.9% on ARC-AGI-3.

Sebastian Raschka reviews OpenAI's GPT-6 Astra, calling it the best model he has used, with disproportionate gains in 3D rendering, animation, and computer use through the Codex/ChatGPT harness. The model scores 99.9% on ARC-AGI-3 versus 7.8% for GPT-5.6 Sol and leads the Artificial Analysis Coding Agent Index, though gains on independent aggregate indices are more incremental. The article also explains looped transformer/recurrent depth architecture rumors, speculation that Astra hides its chain-of-thought reasoning, and recent research insights on the topic.

Hacker News · AIupdated · 3d agofirst · 6d agoModel release 10 sourcesHN 84↑ · 12 comments2

[AINews] Muse Spark 1.3 matches GPT-5.6-Sol, confirming Meta Superintelligence as the newest Frontier Lab, >90% discount for training

Meta's Muse Spark 1.3 reportedly ranks as the world's #3 model, matching frontier models from OpenAI and Anthropic with planned open weights.

The Latent Space AI News roundup leads with Muse Spark 1.3, promised in Zuckerberg's letter, which ranks #3 worldwide per AAII, is slated for open weights, and uses a pricing model over 90% cheaper when users opt in to training. The issue also covers the rumored Gemini 3.8 Flash launch and analysis arguing OpenAI's rumored looped-transformer 'Astra' architecture is a modest tweak rather than a breakthrough. Additional coverage includes ByteDance Seed's HarnessDev harness-evaluation benchmark, a retrieval-invoked actual-use evaluation method, Stanford's revamped agent engineering curricula, and Photon 2.1 adding TTS models and NVIDIA B200 support.

Latent Space · 13d agoModel release1

When Agents Slow Down: Understanding LLM Agents' Test-Time Strategies via Elo-per-token Analysis

Elo-per-token analysis shows LLM agents' marginal gains drop below independent sampling at scale; parallel sessions beat one long session.

The paper proposes Elo-per-token analysis, using a Bradley-Terry model to measure how agent performance scales with token budget on open-ended tasks with continuous scoring. Across four agents and four benchmarks with sessions up to 100M tokens, agents initially convert tokens to Elo faster than independent sampling but eventually slow below the linear-in-log-compute reference. The authors define a scaling inflection point and show that splitting 100M tokens across parallel sessions on FrontierCS Polyomino Packing gains +264 Elo over one long session and +355 over ten short sessions. Human contestants on shared AtCoder Heuristic Contest tasks improve superlinearly, indicating headroom over current agents.

Hugging Face daily papers · 2d agoAI research3· 2 reads

DeepSeek AI Released DeepSeek-V4.1-Flash with 1M Context, FP4 KV Cache, and Cross-Layer Attention Reuse

DeepSeek released open-weight V4.1-Flash, a 552B MoE model with 1M context and FP4 KV cache, beating Opus-5 and GPT-5.6 Sol on agent benchmarks.

DeepSeek-V4.1-Flash is a multimodal Mixture-of-Experts model with a 552B-parameter backbone plus 196B Engram parameters, activating 8B parameters at prefill and 16B at decode, with a 1M-token context window. It introduces a causal encoder-decoder design, Compressed Sparse Attention 2, and FP4 (E2M1) KV cache quantization, cutting global KV cache to 890 bytes per token, about 1/4 of V4-Flash and 437x smaller than V1. Pre-training covered 45T multimodal tokens; the MIT-licensed weights ship on Hugging Face with vLLM and SGLang support. It scores 90.6 on Terminal-Bench 2.1 and 74.2 on DeepSWE v1.1, ahead of Opus-5 and GPT-5.6 Sol.

MarkTechPost · 6d agoModel release1

Benchmarking Qwen3.8 27B quantizations: 4-bit holds up, 1-bit collapses

Quesma benchmarks Qwen3.8 27B quantizations: 4-bit Q4_K_M matches BF16 on key benchmarks while 1-bit collapses to random chance.

Quesma spent roughly $3,000 on Modal GPUs testing Unsloth GGUF quantizations of Qwen3.8 27B across GPQA Diamond, IFBench, and Terminal-Bench 2.1. The 17 GB Q4_K_M quantization matched the 55 GB BF16 model on Terminal-Bench 2.1 and showed little degradation down to 4-bit, while the 2-bit UD-Q2_K_XL dropped noticeably. At 1-bit, scores on GPQA Diamond fell to random-guess levels, with longer reasoning making results worse, and reasoning effort settings significantly affected outcomes.

Hacker News · AI · 7d agoAI researchHN 56↑ · 30 comments1

[AINews] GPT-6 Astra: OpenAI’s biggest LLM launch of all time

OpenAI launched GPT-6 Astra, its new flagship model, claiming state-of-the-art computer use, software engineering, math, and cybersecurity capabilities.

OpenAI launched GPT-6 Astra as its new flagship model, describing it as its most intelligent and aligned model with state-of-the-art computer use, software engineering, and math/science capabilities. Pricing is $10/$50 per 1M input/output tokens standard ($20/$100 fast tier), rolling out first to limited organizations, then ChatGPT Plus/Pro/Business/Enterprise, the API, and AWS. OpenAI claims 99.9% on ARC-AGI-3, 98% on FrontierMath Tier 4, and 100% on ExploitBench. Artificial Analysis scored Astra 67 on the Coding Agent Index and 61 on the Intelligence Index, behind Claude Fable 5.1, and the system card drew attention for reporting decreased chain-of-thought monitorability despite alignment gains.

Latent Space · 12d agoModel release3