WillyDev (@WillyDevNET) header
WillyDev (@WillyDevNET) profile picture

Si Saber No Es Un Derecho, Seguro Será Un Izquierdo. Born to learn ,Trained to Teach. #AI #DL #ALM #Futurist

Bogotá
Joined October 2010
Photo by WillyDev (@WillyDevNET)Photo by WillyDev (@WillyDevNET)Photo by WillyDev (@WillyDevNET)Photo by WillyDev (@WillyDevNET)Photo by WillyDev (@WillyDevNET)Photo by WillyDev (@WillyDevNET)Photo by WillyDev (@WillyDevNET)Photo by WillyDev (@WillyDevNET)Photo by WillyDev (@WillyDevNET)
Looped Transformers are low-key fascinating. First of all, you need to understand what travels around the loop and the 3 core things you set for the architecture. A Looped Transformer runs the same Transformer block (or stack) for multiple passes, feeding each output back in as the next input. ▪️ What travels around this loop is the hidden state: - Token embeddings form the initial hidden representation. - The shared stack updates that representation via attention and feed-forward operations. - Its output goes back to the stack for another pass, using an updated hidden state. Some designs also reuse the input, add time-step embeddings, or run multiple residual streams. - After the chosen number of passes, the output head converts the selected representation into token scores. That's all to win additional computation without an additional set of block weights for every pass. Although the weights stay the same during inference, changing inputs let the same block produce different updates. During decoding it attends via the KV cache, during prefill it updates masked tokens in parallel. A model can loop its entire stack or repeat only a middle stack between ordinary input and output blocks. Note that repeating the stack doesn't require generating a new text token between passes. ▪️ So there are 3 basic things you need to identify for the specific Looped Transformer architecture: 1. The repeated blocks 2. The state they receive 3. The rule that controls their execution - a fixed number of passes, a user-selected count, or an adaptive rule If you want to nerd out on the rest, like adaptive depth, what happens to the KV-cache and hardware, 2026 breakthroughs, and when another pass is worth the cost - we put together a full guide over here: turingpost.com/p/looped-tran…
1
11
1
40
2,590
WillyDev retweeted
Qwen4-Max by @Alibaba_Qwen is the #1 Chinese model on the leaderboard making Alibaba now the #1 Chinese lab and #5 lab overall, just behind Anthropic, Google, OpenAI and Meta. It debuts overall at #14 on Agent Arena with +7.5% net improvement across 14.7k real-world agentic sessions from our global community, real traces on real tasks. This is a major leap over Qwen3.8-Max - Rank: #39 → #14 overall The Chinese frontier on Agent Arena: - #14 Qwen4-Max (Alibaba): +7.5% - #19 Hy4 (Tencent): +5.7% - #21 Kimi K3.1 Pro (Max) (Moonshot): +4.0% - #26 MiMo-V2.6-Pro (Xiaomi): +1.5% - #30 DeepSeek-V4.1-Flash (Max) (DeepSeek): +1.1% Breaking down performance across signals: Qwen4-Max improves over Qwen3.8-Max across all four reported signals, with particularly large gains in user satisfaction and steerability. - #15 Praise vs. Complaint: +13% (best Chinese model) - #13 Confirmed Success: +9% (best Chinese model) - #19 Steerability: +8% - #29 Bash Reliability: +3% By category, it is a top-ranked Chinese model across all 3: - #12 in Work: +9% - #14 in Chat: +8% - #16 in Code: +8% Congrats to the @Alibaba_Qwen team on this release!
📢 Meet Qwen4-Max, your new AI teammate for work, creativity, and discovery! Qwen4-Max is built on a new MoE architecture with native vision and a 1M-token context window. Highlights: - Outperforms Qwen3.8-Max with up to ~50% fewer tokens, while average cost per task drops over 60% in our internal evaluation. - Advances real-world productivity through greater proficiency with professional tools, stronger long-horizon execution, and higher-quality deliverables. - Handles real tasks, including 3D modeling, game development, manufacturing, finance, law, healthcare, and frontier scientific discovery. - $1.5/M tokens for input, $4.5/M tokens for output, and $0.075/M tokens for implicit caching — all lower than Qwen3.8-Max, with implicit caching down by 70%. The API is live on QwenCloud. You can also try Qwen4-Max on Qwen Studio, QwenWork and Qoder. Start building today! 🚀 - Blog: qwen.ai/blog?id=qwen4-max - API: qwencloud.com/models/qwen4-m… - Token Plan: qwencloud.com/pricing/token-… More to come! The open weights of smaller Qwen4 models will be released. Stay tuned. 👀
55
49
49
567
276,645
WillyDev retweeted
🚀 The full DeepSeek App experience, now via API Search, files, links, safety, and voice. The capabilities behind the DeepSeek App are now available through the DeepSeek API. Enable built-in features with a few parameters, or call standalone APIs such as OCR, ASR, and TTS. Build richer conversations or recreate the full DeepSeek App in your own application. Try our open-source frontend demo, with support for importing DeepSeek conversation history. 🔗 Demo: cdn.deepseek.com/chat-demo 📦 Code: github.com/deepseek-ai/chat-… 📖 Details: api-docs.deepseek.com/guides… 1/4
506
1,260
575
12,365
1,182,493
WillyDev retweeted
¿Va a estallar la burbuja de la IA justo ahora que Anthropic va a salir a bolsa? La fuerte caída de las acciones tecnológicas tras las elecciones intermedias en EE. UU. ha puesto en el foco la inminente salida a bolsa de la propietaria de Claude (lgc) p.dw.com/p/5PREf
2
26
1
85
10,076
Never bet AGAINST @elonmusk NEVER GIVE UP!
65
262
20
1,638
44,754
MiniMax H3-Image is the new #1 open weights image model on both AA-Image-Editing v2.1 and AA-Image-T2I v2.1 Overall, H3-Image ranks #13 in Image Editing and #25 in Text to Image. H3-Image Highspeed, a faster version on MiniMax's API only, ranks #14 and #24. Congratulations to @MiniMax_AI on the release! See below for our analysis of both models in the Artificial Analysis Image Arena 🧵
23
17
13
453
36,948
WillyDev retweeted
Gemini 4 Hydrogen (High) by @GoogleDeepMind debuts #5 on Agent Arena with +14.8% net improvement across 6.5k real-world agentic sessions. Hydrogen stands out for its speed and cost efficiency: - Fastest model in the top 10: 5.3 min/task (median), 3x faster than #1 Claude Fable 5.2 and 2.4x faster than GPT-6 Astra - Cost-efficient: $1.6 cost/task (median), 3.7x cheaper than GPT-6 Astra and 4.7x cheaper than Claude Fable 5.1. API token pricing: $1.5/$7.5 input/output per million tokens Notably, it spends more output tokens per task: 41% more than Carbon, 2.8x more than Argon. Strong across key signals: - #2 Steerability: +21% (behind only Fable 5.2 at +33%; ahead of Carbon at +19% and Opus 5.5 at +18%) - #4 Praise vs. Complaint: +32% - #5 Confirmed Success: +15% Ranking by task category: - #3 in Chat: +19% - #7 in Work: +13% - #9 in Code: +12% Hydrogen delivers competitive performance while completing tasks significantly faster and at a fraction of the cost of several leading models. Congrats to the @GoogleDeepMind team on another impressive release!
Meet Gemini 4 Hydrogen: built for rapid coding, multimodal synthesis, and everyday workflows. It finishes tasks 2.5x faster than 3.8 Flash and surpasses Argon on the @ArtificialAnlys Intelligence Index.
42
116
55
1,337
253,596
Meet Gemini 4 Hydrogen: built for rapid coding, multimodal synthesis, and everyday workflows. It finishes tasks 2.5x faster than 3.8 Flash and surpasses Argon on the @ArtificialAnlys Intelligence Index.
471
627
518
7,522
1,572,032
Google's new Gemini 4 Hydrogen has Intelligence just below Claude Sonnet 5.5 and GPT-6.1 Sol, at a third of Sonnet’s cost but ~2.5x that of GPT-6.1 Sol @GoogleDeepMind has released their third new model in six weeks. Gemini 4 Hydrogen introduces an xhigh reasoning level and scores 54 on the Artificial Analysis Intelligence Index, with strengths in visual reasoning and non-hallucination. This places it 1 point behind Claude Sonnet 5.5 (max) and GPT-6.1 Sol (max), at ~32% of Sonnet’s cost but ~2.5x that of GPT-6.1 Sol. With high reasoning, Gemini 4 Hydrogen matches the Intelligence of Gemini 4 Argon for ~19% of the Cost per Task Key takeaways: ➤ The new xhigh reasoning level scores 54 on the Intelligence Index: xhigh is the first reasoning level above high on a Gemini model, and puts Hydrogen 1 point below Claude Sonnet 5.5 (max) and GPT-6.1 Sol (max), and 4 points ahead of Gemini 4 Argon (high). With default (high) reasoning, Gemini 4 Hydrogen scores 51, just behind GPT-6.1 Sol (xhigh, 52) and Claude Opus 5.5 (high, 52) ➤ Strong visual reasoning, presentation and non-hallucination: On GDP.pdf, a visual reasoning benchmark over long documents, it scores 33% all-pass on medium reasoning, the tied #1 score on the benchmark. On AA-Omniscience, Hydrogen has one of the lowest hallucination rates recorded at 8%, compared with 43% for Claude Sonnet 5.5 (max) and 52% for GPT-6.1 Sol (max). On AA-Briefcase, our agentic knowledge work evaluation, graders rate Hydrogen's presentation above other Gemini models, including Carbon ➤ Weaker on agentic terminal work: Hydrogen scores 54% on Terminal-Bench 4.0 and 50% on Terminal-Bench-Science 0.1, covering agentic coding and scientific research workflows in the terminal. Claude Sonnet 5.5 (max) scores 64% and 53%, and GPT-6.1 Sol (max) scores 56% and 58%. xhigh reasoning helps on the science tasks, lifting Terminal-Bench-Science from 41% at high to 50%, but makes no difference on Terminal-Bench 4.0 (54% at both). Science scores are even across domains, from 46% in earth sciences to 53% in physical sciences ➤ Low cost for its intelligence, with high token use at xhigh: Hydrogen is priced at $1.50/$7.50 per million input/output tokens. At xhigh it costs ~32% as much per task as Claude Sonnet 5.5 (max) for near-equal intelligence, but ~2.5x as much as GPT-6.1 Sol (max). xhigh nearly doubles output tokens compared with high for a 3 point gain in Intelligence, and Cost per Task falls from $3.16 at xhigh to $1.87 at high. Medium ($0.80) and low ($0.29) cost less again, but score 41 and 27 on the Intelligence Index ➤ On the Time per Task vs. Intelligence Pareto frontier: Gemini 4 Hydrogen has an output speed of ~200 output tokens per second. With high, medium, and low reasoning effort, Gemini 4 Hydrogen is on the Time per Task vs. Intelligence Pareto frontier, meaning no models are both more intelligent and have a lower Time per Task Key model details: ➤ Context Window: 1M tokens ➤ Multimodality: Text, image, video, and speech input, with text output ➤ Reasoning efforts: Gemini 4 Hydrogen offers four reasoning levels (low, medium, high, and xhigh) with xhigh new in this release ➤ Pricing: $1.50/$7.50 per 1M input/output tokens. Cached input tokens receive a 95% discount ($0.075) Congratulations to the @GoogleDeepMind team on the launch!
106
183
109
2,299
414,332
.@zep_ai introduced Zep Agent Skills, and it’s really worth your attention. It's a procedural memory system that turns verified, successful AI agent runs into reusable Skills. The idea of procedural memory isn't new, but Zep is trying to turn it into a managed service, with a few technically interesting details ↓ First, here's how it works: 1. Tracing: Zep records the agent's actions, tool calls, and outcomes as a Trace. It can also import Traces from Braintrust and LangSmith. 2. Verification: Checks whether the task actually succeeded using independent evidence, not just the agent's claims. 3. Compilation: A Reflector proposes reusable instructions, while a Curator checks them against the evidence. Each Skill field links back to specific Trace events. 4. Admission: Safety is a very important point. Security scans and quality gates determine whether a new Skill version becomes available. 5. Retrieval: Agents find relevant Skills through lexical search, load instructions when needed, and record outcomes that influence future rankings. Then, there are a few features that are particularly interesting and make Zep's system distinct: • Skills aren't tied to one model. A stronger model can solve a task once, and a weaker model can reuse the procedure without retraining. • Zep uses the open Agent Skills format. Each Skill is packaged as a SKILL.md file with YAML frontmatter and optional files, so compatible harnesses like Claude Code and Codex can load it. • Every Skill field needs evidence. The compiler links instructions to specific Trace events, making it possible to inspect where they came from. It can also patch individual fields instead of rewriting the whole Skill. • Skills have a lifecycle. They’re versioned, checked before activation, and ranked partly by verified outcomes. A Skill can even start from one verified success, but remains provisional until it gathers more evidence. The results are also fascinating: - Skills learned from Opus 5.5 improved three weaker models' pass rates on the same tasks by 4.5–18.8 percentage points across 3 benchmarks. - On EnterpriseOps-Gym, all three weaker models using Skills at low reasoning effort beat max effort without them. Most gains came from tasks the stronger model had already solved. Try Zep Agent Skills yourself here, you may find something interesting: getzep.com/zep-agent-skills/…
11
12
57
5,743
WillyDev retweeted
What made Gartner's Top 10 Strategic Technology Trends for 2027, announced at #GartnerSYM Orlando? Discover the technologies and shifts expected to have the biggest impact on business, AI, security and digital resilience: gtnr.it/4icV895
1
8
1
23
2,975
WillyDev retweeted
AI sentiment by country. Notice anything?
397
334
206
3,389
630,153
New research from Microsoft Research demonstrates that reprogramming just a few dormant neurons in an AI agent can create a backdoor that leaves the agent behaving normally until it hits a specific trigger. While the specifics of this threat are novel, the mitigation is not: Know where your model came from, and be smart about its permissions. Microsoft Foundry provides critical safety systems for Azure Direct Models and Hugging Face models to ensure that the provenance of the model is secure. Microsoft researcher @DimitrisPapail explains the measures that model hubs and vendors should take as well as mitigation strategies for enterprises and users. msft.it/6014ahp6e
25
40
15
243
19,255
WillyDev retweeted
Big news: Claude Fable 5.2 (Max) by @AnthropicAI takes #1 in Agent Arena with an unprecedented lead: +30.4% net improvement, about 1.6x the #2 model Opus-5.5, across 4.7k real-world agentic sessions from our global community. It's also #1 in Code, Chat and Work. A major leap over Claude Fable 5.1 (Max) - Rank: #5 → #1, Net improvement: +14.8% → +30.4% - Median cost per task: $7.7 → $7.3 (6% lower). The API pricing $10/$50 → $8/$40 per million input/output tokens (20% lower) vs. The Frontiers - #1 Claude Fable 5.1 (Max): +30.4% | $7.3/task - #2 Claude Opus 5.5 (Max):+19.5%) | $5.5/task - #3 Gemini 4 Carbon (High):+17.0%) | $1.5/task - #6 GPT-6-Astra (Max): +14.5% | $6.0/task Breaking down performance across signals: Fable 5.2 improves over 5.1 across all four reported signals, with particularly large gains in user satisfaction and steerability. - #1 Praise vs. Complaint , +73%, biggest jump vs +41% the next model. - #1 Steerability: +33% - #1 Confirmed Success: +22% - #3 Bash Reliability: +14% #1 across all major categories - #1 Code: +34% - #1 Work: +31% - #1 Chat: +27% - Versus Fable 5.1: Code #6 → #1 (+19.8 %), Work #4 → #1 (+15.8 %), Chat #6 → #1 (+12.0%). Huge congrats to the @AnthropicAI team on this incredible release!
Introducing Claude Fable 5.2 and Claude Mythos 5.2, designed to tackle the most daunting technical challenges. They’re capable collaborators on ambitious projects in many fields, and show improvements in areas like coding, reasoning, and vision.
136
175
118
2,270
414,300
Anthropic has released Claude Fable 5.2, taking the #1 spot on the Artificial Analysis Intelligence Index v5 at 65, led by gains in science, knowledge, and coding At max effort it scores 65, 5 points ahead of Gemini 4 Carbon (high, 60) and 6 ahead of Claude Opus 5.5 (max, 59). Gains are driven by scientific reasoning, knowledge and FrontierCode 2.0, and roughly matching Opus 5.5 on AA-Briefcase and Terminal-Bench 4.0. Priced at $8/$40 per 1M input/output tokens, a 20% reduction from Fable 5.1, Claude Fable 5.2 averages a Cost per Task of $26.5 with max effort, a ~2.6x increase compared to Opus 5.5 (max, $10.0) and ~2x compared to Fable 5.1 (max, $12.99) Key takeaways: ➤ Claude Fable 5.2 leads 5 of the 9 evaluations in the Intelligence Index: At max effort, Fable 5.2 scores 65, ahead of Gemini 4 Carbon (high, 60), Claude Opus 5.5 (max, 59) and GPT-6 Astra (max, 57), and ~15 points above Claude Fable 5.1 (max, 49). It sets the highest scores on HLE-Diamond (80%, previous best 69% from Gemini 4 Carbon), AA-Omniscience v1.1 (67, previous best 54 from Gemini 4 Carbon), Terminal-Bench-Science (73%, previous best 65% from Gemini 4 Carbon), CritPt v2 hard subset (77%, previous best 72% from GPT-6 Astra) and FrontierCode 2.0 (49%, previous best 46% from Claude Opus 5.5) ➤ Less progress in agentic knowledge work and terminal coding: Claude Fable 5.2 improves on Fable 5.1 across capabilities, but sometimes only to Opus 5.5 levels. It reaches 1785 Elo on AA-Briefcase, within the confidence interval of Opus 5.5 (1800) and just behind Claude Sonnet 5.5 (1813). On GDP.pdf it reaches 24% all-pass against 26% for Opus 5.5. On Terminal-Bench 4.0 it scores 60%, level with Opus 5.5 and behind Kimi K3.1 Pro (67%), Gemini 4 Carbon (65%) and Sonnet 5.5 (64%). AutomationBench-AA is an exception, at 76% against 70% for Opus 5.5, narrowly behind Gemini 4 Carbon, 78% ➤ Lower efforts still take leading positions: With high and xhigh effort, Claude Fable 5.2 scores 62, ahead of every other model at any effort, using 98k output tokens per task against 173k for Opus 5.5 (max). At max effort Fable 5.2 costs ~2.6x more than Opus 5.5, but at lower effort settings it can outperform Opus 5.5 at lower cost per task Other model details: ➤ Context window: 1 million tokens with image and text input ➤ Pricing: $8/$40 per 1M input/output tokens; cache writes $10 with 5-minute TTL, $16 with 1-hour TTL, and cache reads $0.20 ➤ Effort settings: five (low, medium, high, xhigh, max). Intelligence Index evaluations were run at all five with Anthropic's default fallback enabled Congratulations to @AnthropicAI on this impressive release!
127
223
128
2,882
266,000
Every technological revolution has remodeled our institutions – how might AI? In @ScienceMagazine, Microsoft researcher @glenweyl and co-author @profjamesevans explore ‘reverse alignment,’ the idea that using neural networks as a model for reshaping institutions might offer a path toward harnessing the benefits of AI while managing its risks. msft.it/6015aAHYP
4
10
1
41
10,614
WillyDev retweeted
Introducing GPT‑6 Cyber: a major advance in security research that pairs stronger capabilities with better model alignment. The model powered the Linux kernel research we announced yesterday, and security teams at companies like Mozilla, Palo Alto Networks, Calif, and Cisco—alongside our own at OpenAI—are already putting it to work. GPT-6 Cyber is available to select Daybreak Red customers and we are expanding access over time. openai.com/index/introducing…
1,322
2,023
1,445
21,528
4,245,565
OpenAI's GPT Image 2.6 takes #1 and #2 on both AA-Image-T2I v2.1 and AA-Image-Editing v2.1. GPT Image 2.6 Flare is almost twice as fast as GPT Image 2.5 Flare with slightly better quality, and GPT Image 2.6 Sunburst takes the top spot with longer generation times OpenAI released GPT Image 2.6 on November 4 in ChatGPT, Codex and the API. The API offers two models, GPT Image 2.6 Flare and GPT Image 2.6 Sunburst, which follow GPT Image 2.5 Flare and Sunburst. Both support Text to Image and Image Editing, and add animated GIF and SVG output. OpenAI added a complexity setting to the API, which controls how much the model plans before rendering. We benchmarked both models at max quality, with complexity at its API default of high. In our testing, GPT Image 2.6 Flare is almost twice as fast as GPT Image 2.5 Flare while delivering slightly better quality. Its median Text to Image generation time fell from about 55 to 30 seconds, and its editing time fell from about 58 to 32 seconds. GPT Image 2.6 Sunburst goes the other way: it takes longer than GPT Image 2.5 Sunburst, at about 2 minutes 15 seconds per image, and takes the #1 spot. Its largest Text to Image gains are in Reasoning, Text Rendering, Lighting and Material, and it is strongest in UIUX Design. Both models cost $30 per 1M image output tokens, the same as GPT Image 2.5. That is about $0.21 per 1024x1024 image at max quality ($211 per 1k images). GPT Image 2.6 is available in ChatGPT and Codex across all tiers, and in the OpenAI API. Congratulations to @OpenAI on the release! See below for our analysis of GPT Image 2.6 in the Artificial Analysis Image Arena 🧵
33
23
23
326
37,595
WillyDev retweeted
🌊 The next wave of disruption won't come from a single technology. Gartner finds that the convergence of advanced AI, next-generation computing and biodigital integration will drive unprecedented change across industries in the years ahead. The question is: Are technology leaders ready? Prepare for what's ahead: gtnr.it/3VeUNd6
3
32
5
109
7,885