Vishal

Anthropic just made Claude for Enterprise chats turn into real agent jobs that run in their own locked-down cloud sandbox and keep working after you close the laptop. Huge for productivity. It turns on by default Nov 16, so your IT team should look now. #EnterpriseAI

Google, JPMorgan and France's digital agency all patched the same bug in their MCP servers, the plugs that connect AI agents to company tools: the agent fetches a link it read somewhere and nobody checks where it goes. Rolling out agents? Ask every tool vendor if they check.

Google's new EmbeddingGemma 2 lets a phone or laptop search photos, videos and voice notes by just describing them, offline and free. For businesses with piles of site photos, call recordings or inspection videos, that's searchable memory without the cloud bill.

Cloudflare open-sourced Clef: a frozen Qwen3.8-27B that skips writing text and just returns calibrated probabilities for typed questions. Clef-flash decides in ~39 ms. Half the agent code out there is a pricey LLM pretending to be an if statement. This is the if statement.

Okay this one's actually useful. NVIDIA dropped Kumo Tabular. Open model. Feed it labeled rows + unlabeled rows. One forward pass. Predictions out. No training job. No feature store pilgrimage. No three-week AutoML side quest just to ask "will this account churn." We've treated every tabular question like a mini research project for ~20 years. Half the simple predictive asks never leave the backlog because the ceremony costs more than the answer. This flips it. Table + question. Model pretrained on millions of messy synthetic tables. Your labeled slice is the prompt. 28M–215M params. Small enough to actually run. Skeptical until you try it on your data. Distribution shift still bites. Leakage still bites. No free lunch. But if it holds up, a big chunk of classical ML just went from "project" to "API call." Weekend test: same churn table, same split, Kumo vs your GBT. Plot accuracy vs wall clock vs engineering hours. The number that matters isn't AUC. It's time-to-first-honest-answer. Also, Kumo means cloud. Tables. Nice.

In the age of NLP why do you make use of Chat menus? @amazonIN

Dear @OpenAI, your new app has a very noobishly designed UI! My projects are lost in the desktop app. And why on earth would you remove the Record Meeting button? That was the only reason why I had signed up for the plan that I was on! Please roll back!

Meta just released TRIBE v2 ai.meta.com/blog/tribe-v2-br… This is not just another AI model. It predicts brain activity from real-world inputs. Video. Audio. Language. All together. Not narrow. Not siloed. Closer to modeling cognition itself. Trained on large-scale brain data. Generalizes across humans. That’s the key. We are moving from AI that processes data to AI that approximates how humans perceive reality. Implications are serious: • Brain-computer interfaces get real • Neurological disorders become more tractable • Human-level AI becomes less theoretical If you can model perception you can reshape intelligence. – at Pune, India

Let's face it: knowing square roots is less important now with mobile calculators, MS Excel and now AI. Everything in your pocket, nothing necessarily in your mind. Let’s give kids a childhood, not just study hours. The old education system needs a rethink. #EducationReform #LetKidsBeKids #AIinEducation #ParentingWisdom #FutureLearning #RethinkEducation"

MIT study wants you to believe it shattered the AI hype: Smartly creates a news that 95% of generative AI projects are failing, sparking tech bubble jitters Well, sample size makes a difference, MIT(but how would you know). It's like asking Rikshaw drivers about how beneficial Uber was for their business and do you foresee that having impact on their future. For the survey they interviewed: 150 business leaders across 52 companies.(What companies?) 350 employees(Must be who's jobs are at stake) An examination of over 300 public AI deployments.(Must be wannabe Unicorn startups) Well, AI is bad for MIT's business and that is for sure!

Anyone can throw blueberries on a cheesecake. What makes it special is the recipe, the craft, the time. Same with AI products. Well, fast is fragile. What’s built in minutes can be replaced in seconds. Depth outlives speed

Bought the #MetaAI Skyler glasses. Thought I was stepping into the future. Turns out I just paid $300+ to check WhatsApp… hands-free. 🫠 No Outlook. No calendar alerts. No dev SDK. No real apps beyond Zuck’s backyard. Tried wearing them in a meeting; notification audio was distracting, and watching back the recording made me mildly nauseous. 🤢 Tried using them like a GoPro? Hilarious. Not IP-rated, so forget rain and don't you dare drop them while playing. Battery barely lasts an hour in capture mode, and video recording tops out at 3 minutes. Basically a tech demo with nice hinges. Only genuinely useful things: "Summarize" and "Snap." Which, let’s be honest, is just fancy talk for “take a photo” and “give me a TLDR because I wasn’t paying attention.” Can’t even pretend I’m being productive with these. Just expensive sunglasses with an AI friend who lives in denial. Still wearing them, though. Because I paid for them. Waiting for #Google to drop something that doesn’t feel like a beta test in disguise. 🥲👓

The Sun captured with a Samsung Galaxy S24 Ultra...no filter, just raw solar brilliance revealing a distinct sunspot.