AI Developer Code @AIDeveloperCode

@NVIDIAAI @david_nix How you buy it man !

Agent Harness is just a long running ReAct Loop !

Dario is wrong. He knows absolutely nothing about the effects of technological revolutions on the labor market. Don't listen to him, Sam, Yoshua, Geoff, or me on this topic. Listen to economists who have spent their career studying this, like @Ph_Aghion , @erikbryn , @DAcemogluMIT , @amcafee , @davidautor

TFTC @TFTC21

2 months ago

Anthropic CEO Dario Amodei: “50% of all tech jobs, entry-level lawyers, consultants, and finance professionals will be completely wiped out within 1–5 years.”

1K 1K 6K 8.5M 3K

@GeminiApp + YOLO > @claudeai

Thank you for reading and sharing

Ever been locked out of your Facebook or Instagram and felt that gut-drop panic? That moment just got a lot less stressful. Meta is rolling out a centralized support hub and AI-powered improvements that make help easier to find and account recovery faster and more reliable. Here’s what’s new: • A single support hub inside Facebook and Instagram to report issues, start recovery, and find answers. 24/7 access, plus an AI assistant pilot for quick, personalized guidance starting on Facebook. • AI-driven safety upgrades that Meta says reduced new account hacks by more than 30% globally and increased the relative success rate of hacked-account recovery in the US and Canada by more than 30% over the past year. • Smarter, adaptive recovery flows, including optional selfie video verification when other signals aren’t enough. • Easier access to protections you can enable today: Security Checkup, two-factor authentication (2FA), and passkeys. Why it matters: Your accounts hold years of photos, messages, and connections. When something goes wrong, you need clear steps that work the first time. The new hub is designed to reduce friction, cut dead ends, and guide you to the right fix—fast. Do these 5 things now: 1) Run Security Checkup on Facebook. It takes minutes and surfaces simple fixes. 2) Turn on 2FA with an authentication app for both Facebook and Instagram. Add backup codes and, if needed, multiple devices on Instagram. 3) Add a passkey in Accounts Center for passwordless sign-in using your device’s face, fingerprint, or PIN. 4) Review active sessions and trusted devices; sign out of any you don’t recognize. 5) Learn where the new support hub lives in your app so you can find it fast when you need it. Real-world scenarios: • Lost access to your email or phone? Start recovery in the hub and choose “I can’t access this email or phone.” The flow adapts and may offer selfie video if needed. • Attacker turned on 2FA to lock you out? Use the hacked-account flow; the system will look for trusted-device signals and known locations to verify you instead of relying on the attacker’s 2FA. • Traveling and got a risky-login alert? If it’s not you, change your password immediately and run Security Checkup. Passkeys vs. 2FA: • 2FA adds a second step after your password. • Passkeys replace the password altogether with your device’s built-in authentication, making sign-in faster and resistant to phishing. Security mindset: • Never share codes, screens, or payments with anyone claiming to be “Meta Support.” Use in-app help and the official recovery paths only. • Keep your apps updated; many protections are delivered silently in the background. Bottom line: Getting help on Facebook and Instagram hasn’t always been straightforward. The new support hub, the AI assistant pilot, and smarter recovery flows are meaningful steps forward. Turn on 2FA, add a passkey, run Security Checkup, and keep recovery paths handy. Those small steps today can save hours when it matters most. 🔐📲 #Facebook #Instagram #Meta #Security #AccountRecovery #Passkeys #2FA #Privacy #Cybersecurity #MetaAI

Thank you for reading and sharing

AI Week in Review: Nov 15-22, 2025 **Key Highlights to Act On:** 1) **UK’s Strategic Moves** - The DSIT released the AI for Science Strategy featuring 15 actions to accelerate research, upgrade national labs, and allocate compute through the AI Research Resource (AIRR). It balances ambition with accountability. #UKAI #SciencePolicy 2) **New National AI Data Facility** - The University of Bristol has announced a National AI Data Facility next to Isambard-AI, enhancing rapid access to data alongside robust computational capabilities. #IsambardAI #DataInfrastructure 3) **AI Growth Zones in Wales** - The announcement of new AI Growth Zones in North and South Wales aims to foster job creation and link AI investment with regional strengths, encompassing energy sectors. #RegionalGrowth #TechJobs 4) **Moody’s Economic Outlook** - Moody’s affirmed the UK’s Aa3 rating with a stable outlook, supporting favorable financing conditions for large-scale AI initiatives in the coming years. #Ratings #UKEconomy 5) **Widespread AI Adoption** - McKinsey’s recent survey indicates a significant increase in AI use across organizations. Stanford’s AI Index reports a dramatic reduction in costs for running advanced models, making AI more accessible for teams. #AIAdoption #AICosts 6) **OpenAI-Foxconn Collaboration** - OpenAI and Foxconn are working together to co-design AI hardware components in the US, targeting supply chain resilience amid growing demand. #AIHardware #Manufacturing 7) **Synthetic Media Regulations in India** - India’s proposal for labeling AI-generated content highlights growing efforts to ensure content integrity and combat disinformation. #AIProvenance #Deepfakes 8) **Significant Research** - University of Surrey initiatives unveil brain-inspired models for energy-efficient AI and techniques for predicting osteoarthritis progression from X-ray imaging—both underscoring responsible AI development. #ResponsibleAI #HealthcareAI **Next Steps for Leaders:** - Reassess total costs for inference-heavy workloads; costs have shifted dramatically. - Treat data with the same rigor as product development; document lineage and access as required. - Evaluate supply chain vulnerabilities, including energy dependencies. - Keep an eye on AIRR allocation opportunities and UK data initiatives. **The Takeaway**: Focused infrastructure is pivotal. Align falling costs with dependable data and agile hardware planning to ensure readiness for the future of AI in 2026. #AI #UKTech #Compute #Data #Policy #OpenAI #Foxconn #StanfordAIIndex #IsambardAI #Innovation #Engineering #Research #Startups #EnterpriseAI #GenAI #AIAgents #Sustainability #Security #Provenance #LLM #AIWeek #Roundup 🤖📉📊⚙️💡🇬🇧

Thank you for reading and sharing

I spent a week trying 5 AI browsers so you don’t have to. Here’s the honest 2025 verdict. **The contenders:** - Chrome with Gemini - Microsoft Edge with Copilot Mode - ChatGPT Atlas - Perplexity Comet - Dia by The Browser Company **What worked:** - **Read-and-reason workflows:** Faster summaries, outlines, and multi-tab comparisons made tasks less tedious. - **Side-by-side help:** Panels in Edge, Chrome, and Atlas facilitated smooth follow-up actions. - **Study and note-making:** Dia's Skills and Atlas highlights excel for students and researchers. **What didn’t:** - **Hands-off automation:** Agents require supervision and can stall or mis-click. - **Aesthetic matters:** Shopping still needs human judgment for taste and context. - **Security comfort:** Be aware of prompt injection risks; keep agents on a tight leash. **Quick picks:** - **Chrome with Gemini:** For fast summaries without a new setup. - **Edge with Copilot Mode:** A solid sidekick on Windows with helpful features. - **ChatGPT Atlas:** Best if you’re already in the ChatGPT ecosystem on Mac. - **Perplexity Comet:** Ideal for a more agentic approach, but stay vigilant. - **Dia:** A daily driver with a friendly chat feature perfect for study tasks. **Safety tips:** - Default agents to read-only mode. - Use separate profiles for sensitive work. - Avoid assigning banking or health tasks to agents. - Keep memory limited and regularly cleared. - Review links, forms, and drafts before submitting. **Bottom line:** AI browsers are powerful tools, not autopilots. Leverage them for reducing friction in reading and comparing—always keeping your judgment in the loop for tasks that involve clicking and typing for you. #AI #browser #Gemini #Copilot #ChatGPT #Atlas #Comet #Dia #Productivity #Security #PromptInjection #AgenticAI #Research #EdTech #Windows #macOS #Android #Chrome #Edge #LLM #GenAI 🚀🧠🔒💻📚🧭🛠️

Thank you for reading and sharing

The biggest AI story of late 2025 isn’t a new model or chip. It’s a power struggle over who gets to write the rules. Is it Washington with a single, unified standard - or the states that moved first to protect residents from deepfakes, discrimination, and safety failures? Here’s what matters now: 1) Preemption push vs states’ rights. Reporting shows renewed attempts to curb state AI laws via must-pass bills and even a draft executive order that would create a DOJ “AI Litigation Task Force” and nudge FCC/FTC toward national standards. That would set up a federal vs state showdown and immediate court fights. ⚖️ 2) States acted early. California’s SB 53 is law with transparency and incident reporting for frontier AI developers. Colorado’s risk-based AI Act kicks in during 2026 after an extension. Texas adopted a narrower regime with a statewide sandbox and preemption of local rules. New York’s RAISE Act cleared the legislature and awaits the governor’s call. 🗺️ 3) The Senate’s 99-1 clue. In July, senators voted 99-1 to strip a 10-year moratorium on state AI laws. The message: don’t bench the states without a serious federal plan. ✅ 4) Political money enters the chat. Pro-AI groups say they’ve raised $100M+ to back candidates favoring a single national standard and to oppose tougher state rules. Whether you see that as pro-innovation or self-interest, expect more ads and sharper rhetoric in 2026. 💸 5) The White House’s other move: the Genesis Mission. DOE will lead a national AI-for-science platform to accelerate discovery, part of a broader argument for U.S. leadership that also fuels calls for a unified rulebook. 🔬 What smart teams are doing now: - Map AI uses and jurisdictions - Align governance to NIST AI RMF - Test for bias and robustness in high-stakes uses - Build incident reporting playbooks - Prepare clear consumer disclosures - Monitor NY, TX, CO rulemaking and any federal riders The likely endgame: a federal baseline that sets clear, national guardrails - with room for states to fill gaps. Until then, plan for a mosaic. The companies that operationalize risk management, documentation, human oversight, testing, and transparency will be ready no matter who wins the jurisdiction fight. #AI #Policy #Governance #TechLaw #Compliance #NDAA #ExecutiveOrder #SB53 #ColoradoAI #TexasAI #RAISEAct #NIST 🤖📜🏛️

Thank you for reading and sharing

The next AI revolution is upon us. Google has released Gemini 3, embedding it into Search, the Gemini app, and developer tools. In an immediate response, OpenAI has reportedly declared a "code red" to enhance the ChatGPT experience. This rivalry is set to influence how billions of us search, work, code, and learn in 2026. Key developments include: - Gemini 3 integrates seamlessly with Google services, shifting the AI landscape towards deeper thinking and autonomous actions. - OpenAI is prioritizing ChatGPT's speed and personalization, focusing on creating a noticeably better user experience. - Anthropic is pushing ahead in the enterprise sector with its Claude 4.5 models focused on efficiency. Looking forward: - Distribution plays a crucial role. A shift in user behavior with AI Mode or Gemini could give an edge to Google. Conversely, if OpenAI enhances its assistant features, it can close the gap. - Agent reliability will be a competitive factor. The platform that efficiently manages complex tasks with minimal user oversight will likely capture the developer community's support. - Trust and safety are now competitive advantages, with behavior and transparency shaping user loyalty. What does this mean for you? - Consumers will see enhanced AI capabilities in everyday tools. Expect improved responses from Google Search and ChatGPT. - Professionals will benefit from AI's better handling of complex tasks. Experiment with both platforms for your specific needs. - Leaders should link AI implementation to clear results and maintain a flexible multi-vendor approach. **Hype vs. Reality:** Case studies matter, but sustained daily value across diverse user scenarios will determine success. Be sure to test both platforms in practical applications. **Bottom Line:** The competition is effective. Google’s proactive approach with Gemini 3 is reshaping standards, while OpenAI is enhancing ChatGPT. Anthropic continues to elevate enterprise-specific reliability. In 2026, success will be about trust, safety, and execution. #OpenAI #Google #Gemini3 #ChatGPT #GPT5 #Anthropic #AI #ArtificialIntelligence #AIAgents #Search #Productivity #Developers #TechStrategy #AIFuture 🤖🚀💡

Thank you for reading and sharing

FLUX.2 has just been released - engineered for actual creative workflows. Here’s a brief overview to help you assess its fit for your toolkit. **What it is:** - A new image generation family from Black Forest Labs tailored for production workflows. - Generates photorealistic images, ensures better prompt adherence, and delivers reliable text rendering. - Features multi-reference control for maintaining character or product consistency across campaigns. - An open-weight dev model for local runs, ideal for research and prototyping. **Why it matters:** - Consistency at scale is critical for many teams. FLUX.2 allows referencing up to 10 images at once, maintaining identity and style throughout. - With support for outputs up to 4MP and any aspect ratio on hosted systems, it produces images that resemble real photos, ideal for marketing, e-commerce, and design. **Model family:** - FLUX.2 Pro: Offers the highest quality and speed on BFL’s endpoints. - FLUX.2 Flex: Developer-focused controls for balancing quality, speed, and typography. - FLUX.2 Dev: A 32B open-weight model that incorporates both generation and editing features. **Running it:** - Local runs require significant VRAM—without offloading, models can exceed 80 GB; with CPU offloading and 4-bit quantization, 24 GB GPUs become viable. - Group offloading is possible on 8 GB GPUs if paired with enough system RAM. - Live support with Diffusers, including Flash Attention 3 options on newer NVIDIA GPUs. **Where it excels:** - Marketing and advertising: Consistent branding across campaigns and product visuals. - Product visualization: Producing photoreal renders with appropriate lighting. - UI and design: Comprehensive mockups with readable text and structured layouts. - E-commerce: Large-scale product photography with coherent lifestyle imagery. **What changed under the hood:** - New autoencoder redefines the latent space for enhanced quality. - Optimized transformer backbone for heightened efficiency and fidelity. - Implements single-encoder text conditioning within the open pipeline. **Licensing tips:** - FLUX.2 Dev is not for commercial usage by default—be sure to check the license agreements. - FLUX.2 VAE is Apache 2.0 licensed for reuse. - For commercial initiatives and scale, consider using the hosted Pro or Flex services. **Bottom line:** FLUX.2 advances image generation towards practical and production-ready workflows, boasting improved photoreal quality, better typography, and multi-reference consistency. This tool is certainly worth exploring for visual creators, product teams, and campaign managers. #FLUX2 #GenAI #AIimage #Diffusers #HuggingFace #OpenWeights #Mistral #LoRA #ProductDesign #Ecommerce #MarketingTech #AIGenerative #VisionAI 🚀🖼️✨📸

Thank you for reading and sharing

Curious about how some chat interfaces remain so slick, even under heavy load? The magic lies in a concept called continuous batching. To simplify: continuous batching keeps GPUs firing on all cylinders by blending two types of tasks— 1) **Prefill** involves the model interpreting the full prompt to predict the next token, which is a resource-heavy initial pass. 2) **Decode** steps are carried out one at a time after the first token appears in a continuous stream. Here's the formula modern engines apply: - **KV Caching:** Compute Key/Value states once during prefill, allowing them to be reused in subsequent decode steps for efficiency. - **Chunked Prefill:** Divide lengthy prompts into manageable segments that fit into memory alongside decoding tasks. - **Ragged Batching + Dynamic Scheduling:** Combine tokens from various requests into a single attention pass without padding, with the ability to seamlessly swap requests as needed. **Why this approach is effective:** - No wasted computations on padded tokens. - High GPU utilization rates across varying request times. - Flexibility in choosing between low latency and high throughput based on your needs. **Real-world applications:** - **Hugging Face TGI** provides a built-in continuous batching feature alongside optimizations like Flash Attention and PagedAttention for enhanced KV memory management. - **vLLM** has popularized the PagedAttention technique to minimize KV cache fragmentation, allowing for greater concurrency and better overall throughput. - Encoder-decoder models utilize similar principles under what NVIDIA describes as in-flight batching, effectively managing both self- and cross-attention caches. **Implementation tips:** - Prioritize decode tokens for snappy response times. - Use prefill chunks strategically to fill remaining capacity without impeding decoding work. - Define a max tokens-per-step budget based on your GPU memory capacity. - Monitor request lengths and set limits to retain effective batching. - Consider implementing page-style KV memory managers to curb fragmentation issues. **In summary**, always compute using actual tokens, never fall back to padding. Apply this principle at scale, and your LLM service will deliver fluid experiences, even when demand peaks. #LLM #AI #Inference #MLOps #Systems #GPU #GenAI #Optimization #vLLM #TGI #FlashAttention #PagedAttention 🚀🧠⚙️

Thank you for reading and sharing

Here’s your clear, verified recap of the week in tech and AI (November 17–24, 2025): • Google launches Gemini 3 across Search (AI Mode), the Gemini app, and developer platforms. It boosts reasoning, multimodal understanding, and agentic capabilities—and debuts an agent-first dev environment called Google Antigravity for building task‑oriented workflows. Some outlets also note an upgraded image model surfaced for users. #Gemini3 #GoogleAI #AgenticAI 🔧🧠 • xAI rolls out Grok 4.1 on web and mobile, emphasizing fewer hallucinations, better creative writing, and improved emotional understanding. New API options enable faster tool‑calling agents and an Agent Tools API for autonomous workflows. #Grok #xAI 🤖⚡ • Adobe announces a $1.9B deal to acquire Semrush, aiming to unify classic SEO with generative engine optimization (GEO) so brands can track visibility in search and inside AI assistants. #Adobe #Semrush #MarketingTech 💼📈 • Cloudflare suffers a significant outage that knocks major sites and services offline, including X and ChatGPT. The incident was resolved, but it’s a reminder to build for resilience and graceful degradation. #Cloudflare #Infra 🌐🚧 • OpenAI adds group chats to ChatGPT—moving AI from solo prompts to shared planning and collaboration. Rolling out globally across tiers. #OpenAI #ChatGPT 🧑‍🤝‍🧑💬 • Amazon Music x TikTok launch Share to TikTok—post songs, playlists, and Amazon Music Insights directly to TikTok. #AmazonMusic #TikTok 🎵📲 • Salesforce expands Agentforce, adding retail and commerce skills (guided shopping, order management, agentic selling into AI channels like ChatGPT) while keeping merchant data in‑house. #Salesforce #Agentforce 🛍️🤝 • Jeff Bezos returns to an executive role as co‑CEO of Project Prometheus, a new AI startup reportedly funded at $6.2B to bring AI to engineering and manufacturing. #Bezos #ProjectPrometheus 🚀🏗️ • Berkshire Hathaway reveals a multibillion‑dollar Alphabet stake, signaling confidence in Google’s AI trajectory. #Alphabet #BerkshireHathaway 💰📊 Consumer features getting real: – Google expands AI‑powered trip planning: describe your trip, get an itinerary in Canvas, and use agentic booking for restaurants and more. Flight Deals expands globally to 200+ countries. #TravelTech ✈️🗺️ Trending tools to try: – MagicSchool: AI for educators (lesson plans, assessments, student support). – GPTZero: widely used AI detection (treat detections as signals, not verdicts). – Originality.ai: AI + plagiarism checks for editors and publishers. – Meshy: text/image‑to‑3D asset generation. – Kling AI: fast short‑form text‑to‑video with cinematic results. – PolyAI: lifelike enterprise voice agents for customer service. #EdTech #ContentOps #3D #VideoAI #VoiceAI 🧑‍🏫📰📦🎬📞 Why this week matters: 1) Models are becoming agents. Gemini 3 and Grok 4.1 emphasize planning, tool use, and autonomy. 2) Discovery is shifting beyond SEO. Adobe + Semrush points to an AI‑assistant future where GEO matters. 3) Resilience is a must‑have. The Cloudflare outage shows how dependencies ripple—plan for failover and clear comms. 4) Collaboration features will pull AI into team workflows (OpenAI group chats). 5) Everyday utility is here. Google’s AI trip planning is a practical, mainstream agentic use case. Action items: • Pilot agentic workflows in one or two business processes. • Start measuring brand visibility across both SEO and LLM surfaces. • Rehearse your outage playbook—graceful degradation, status updates, and fallbacks. • Establish collaboration norms when AI joins team chats. • For creators and marketers, test image/video/voice tools to see where they fit your pipeline. It’s not just about bigger models anymore—it’s about dependable systems that plan, act, and explain their work. Onward. 🚀 #AI #TechNews #Gemini3 #Grok #OpenAI #Cloudflare #Adobe #Semrush #Salesforce #Agentforce #ProjectPrometheus #Alphabet #SEO #GEO #AgenticAI #Search #TravelTech #EdTech #VoiceAI #VideoAI #3D