This reply hit a nerve. 14K views under Yann LeCun's post about Dario's '50% of jobs wiped out' prediction. The gap between what AI can do in a demo and what enterprises can actually ship is the most expensive blind spot in tech right now. Most 'AI transformation' is cost-cutting rebranded. Only one version, actually upskilling your workforce to be 2-3x productive, creates real ROI. The rest is earnings call theater.

Hossam Alaa al-Din Abd al-Jabbar Walad Ali (38) was killed, and two young men were seriously injured by Israeli occupation forces’ gunfire near the town of Bani Na’im, north of Hebron, on Monday morning.

“A mother described to me that she had prepared her child’s snack and placed it in her backpack, kissed her face, and sent her to school. What was returned to her was her child’s backpack with an uneaten sandwich.” dropsitenews.com/p/minab-sch…

Iranians deserve safe air travel just like any other people on earth. Instead, the US government is using pure bullying to cut off safe transit for an entire nation of 95 million citizens. Weaponizing civil aviation against civilians is unacceptable. #SafeSkiesForIran #AirApartheidAgainstIran

So @SecScottBessent is stating that he will "shut down" all Iranian airlines on September 23. What he hasn't explained is how this advances U.S. national security. What we know is that this is an attack on Iranian healthcare supply chains. Over 60% of Iranian medical imports arrive by air. Of this total, around one-third are “medicaments containing hormones,” a category which includes insulin for treatment of diabetes and certain cancer treatments. Just over 10% are vaccines, both for human and veterinarian use. Another 6% includes blood products used to treat infections or support clotting. Bessent's move threatens the health of ordinary Iranians. To what end? bourseandbazaar-substack.org…

Tabular FMs scaling past TabPFN's row limits is the real milestone here. But for scientific and risk-sensitive tables, deployment isn't decided by benchmark accuracy, it's whether calibration holds under distribution shift. Confidently wrong is the failure mode leaderboards miss!

Voice agents are going to make infra bugs look like model bugs. WebRTC drops audio to protect latency, then the model gets blamed for missing context! The eval should start at the packet boundary, not the prompt template.

GPT-5.5 and DeepSeek V4 dropped on the same day! If you’re an enterprise CTO, the message is clear: the model you chose last quarter is already obsolete. The real question isn’t which model is best. It’s whether your AI systems can survive model swaps without breaking your eval pipeline, your permissions layer, and your audit trail. The teams treating model selection as a one-time decision are building on sand!

Anthropic dropped Opus 4.7 yesterday. Reaction is split. Benchmarks: real step forward. 87.6% on SWE-bench Verified, strong agentic improvements, better vision. But developers are louder about what they lost. Updated tokenizer quietly inflates token counts up to 35%. Same price, fewer prompts per dollar. I've been running Claude in production pipelines for months. The agentic improvements are real. Fewer tool errors, proactive self-verification before surfacing results. That matters more than benchmark points if you're running multi-step automation. The bigger story is what Anthropic isn't shipping. They openly concede Mythos scores 93.9% on SWE-bench vs Opus 4.7's 87.6% and stays locked for safety. The public frontier and the actual frontier are now two different things.

Feels like the world is getting used to wars and “peace deals” being announced via tweets…like it’s all just a click of a button. That’s not how reality works. On the ground, this ongoing war is still disrupting energy flows and has shut in around 11.5M bpd of production. Yes, oil prices are down, but let’s be clear, nothing fundamental has actually changed. #OOTT

Respectfully disagree with @HFI_Research on 0%. Saudi dominance in OPEC is a post-1979 accident, not a geological constant. Pre-revolution, Iran was co-equal. That history matters when we ask who really holds structural power in the Gulf. 38 days of evidence so far. Saudi exports disrupted. Infrastructure struck. A 35-nation Hormuz meeting that produced zero military commitments. You can call that strategic restraint or you can call it incapacity. Either way the result is the same: no action. Kpler is tracking 32 vessels through a two-tier transit system since March 30. Early signal, small sample, not proof of anything permanent. But Iran is clearly testing whether selective transit control is sustainable. Worth watching closely. The real counter-scenario is a Saudi price war. Flood the market, collapse prices, punish Iran economically. I'd put that around 15 to 20 percent probability over six months. The problem is Saudi's own budget needs $90+ oil to balance, so that move is self-harm. And physically reopening the strait requires US 5th Fleet escalation that nobody seems eager to trigger. The better question isn't "will Saudi allow this forever." It's how long persistent friction can sustain a risk premium before a credible response materializes, and what that response actually costs everyone involved. 0% assumes a binary: toll booth or no toll booth. Markets are pricing something messier. Elevated insurance, selective transit, structural uncertainty. That doesn't require Saudi agreement. It just requires time.

RT @ForParastou: پرستو فروهر: در این تاریکی، پای ایران و آزادی بایستیم radiozamaneh.com/884961