All Activity

Amodei addressed the issue in a blog post on Monday, just days after Nvidia, Microsoft, Meta, Palantir and several other technology companies signed a letter urging US policymakers not to introduce "premature restrictions" on open-weight AI models (these are models that users can download, modify and run on their own infrastructure). View the full article

OpenAI's CEO is keen to test its inaugural hardware in South Korea, aimed for release by the first quarter of next year. This innovative screenless smart speaker is designed to be a versatile, mobile AI assistant for households. However, potential legal hurdles and trademark challenges could affect its launch. With South Korea's robust infrastructure and quick adoption of new tech, the market is ripe for such advancements. View the full article

European Union's new artificial intelligence rules mandate labeling of AI-generated content. These regulations aim to ensure users can distinguish real online material from synthetic creations. Companies face significant penalties for non-compliance with these transparency requirements. Major technology firms are already implementing their own labeling systems in anticipation. The EU's comprehensive law seeks to preserve public trust in digital information. View the full article

The United States could face "massive capital destruction" from its artificial intelligence (AI) investment boom as Chinese open-source AI models increasingly challenge the dominance and profitability of US hyperscalers, according to a report by Jefferies. View the full article

1. rajan.arora2000 Position: View B (Continue pursuing improvement / retain the routine work) — stated unqualified, with one derived exception (keep only the "Teaching Share"). Specific Example: An unusually deep documented portfolio — radiology (Hinton's 2016 "stop training radiologists" call; MD-senior applications falling to 3.8% by 2015; residency slots 1,090→1,449; avg comp ~$571k; Mayo running 250+ AI models with 400+ radiologists, +55%); US air traffic control (~11,000 CPCs vs 12,563 required; 2–6 yr certification; ~2% applicant completion; $200M/2.2M hrs overtime in 2024); Klarna (700 AI agents, 2.3M chats, 5,000→3,500 headcount, partial reversal); Vogtle 3&4 ($14B→$36.8B, 8-yr delay); the German dual system (€18k/€12k/€6k per apprentice-year; 74% retention); Bastani et al. 2024 (−17% unassisted, n≈1,000); Shen & Tamkin 2026; Post Office Horizon; the Statute of Artificers 1563. Roughly 19 precedents across 11 sectors, with cited papers, case citations, dates and figures. Reasoning Quality: Exceptional — the argument reframes the decision (output vs. training-system pricing), builds original quantitative models (the "Last-Third Problem" reducing the contested figure from $3M to ~$0.9M; break-even premium tables), steelmans the opposing view, labels its own confounds and a deliberate negative control (TAR/Da Silva Moore), and ties every example explicitly to a mechanism. The main weakness is length and density that risks burying the decision, but the rigor and evidentiary grounding are far beyond threshold. 2. GoutamNamata Position: View A (Accept the AI's recommendation / give routine work to AI) — clearly stated. Specific Example: Microsoft using GitHub Copilot to automate routine coding (boilerplate, unit testing, documentation, debugging), freeing junior engineers for reviews, design, and security work. Reasoning Quality: Competent — the logic is coherent (reinvest savings into structured learning; expose juniors to complex work earlier), and the position is clear. But the example is essentially a name-drop of a known tool with no documented outcomes: no timelines, no financial figures, no cited productivity data or sources, and no evidence that the training-substitution actually worked. 3. Sivakumar_Raghava_06OP Position: View B — clearly stated. Specific Example: Radiology (AI reading routine chest X-rays/CT scans; residents losing the "grind"; a 7–10 year retirement gap) and a secondary accounting-pipeline illustration, presented mostly as a hypothetical trade-off table built from the scenario's own $3M figure. Reasoning Quality: Good — well-structured, with a clear trade-off table and a solid articulation of tacit-skill development and the "who catches AI's mistakes" problem. The weakness is decisive for approval: the radiology example is treated as a generic illustration rather than a documented case — no Hinton forecast, no application-rate figures, no shortage numbers, no sources. 4. Suhail_J_CaJq Position: View B — clearly stated. Specific Example: AWS/Amazon — CEO Matt Garman's on-record remark that replacing junior developers with AI is "one of the dumbest things I've ever heard," AWS hiring ~11,000 interns and recent graduates this year, set against Amazon cutting ~14,000 corporate jobs the prior fall, tied to the scenario's "60% gone in 7 years" cliff. Reasoning Quality: High quality — connects a specific, quoted, figure-backed corporate decision directly to the scenario's mechanics (an AI company protecting its own junior pipeline precisely because its tools automate that training ground), and makes a genuine concession to View A (sequencing/AI-as-tutor) that strengthens rather than weakens the position. Slightly narrower single-example base than the top entries, but grounded and current. 5. Sreenivas chaitanya_Guttarlapalli_t6cZ Position: View B — clearly stated. Specific Example: None — refers generally to a "junior team" replacing a "senior team" over five years and to "live systems" where AI may fail, with no named organization, sector case, or figures. Reasoning Quality: Reasonable — the core intuition (short-term savings vs. long-term pipeline investment) is sound but asserted rather than demonstrated. It is a brief, generic argument with no grounded example. 6. Mukul_Nagpal_oazD Position: View B — clearly stated. Specific Example: Radiology (AI reviewing routine scans; experienced radiologists still needed for unusual cases and clinical context; juniors missing gradual learning), framed around the scenario's "60% in 7 years" figure. Reasoning Quality: Good — thoughtful reframing of the decision as a workforce-sustainability rather than cost issue, and a clear articulation of the "AI-generated output but nobody to judge it" risk. However, the radiology example is used generically, with no documented outcomes, figures, or sources. 7. kanchan_vishwakarma_cwnh Position: Unclear — offers a general "Learn First, Automate Later" personal framework (70/20/10 rule, three-phase approach) rather than taking a position on the organization's View A vs. View B decision. Specific Example: None — lists skill categories (SQL, SAP, Excel, Six Sigma) and prompt templates, but no real-world organizational case. Reasoning Quality: Reasonable as generic advice, but it does not engage the actual scenario decision, states no clear View A/B position, and provides no grounded example. 8. Savio Dsouza Position: View B — clearly stated. Specific Example: Radiology (Hinton's 2016 forecast; application rates falling from 5–7% to a 25-year low of 3.8%; a current shortage of ~1,500 radiologists potentially widening to ~3,100; hospitals closing outpatient imaging centers; only 29 new first-year diagnostic radiology slots added 2021–2025 due to federal funding caps; Siemens Healthineers referenced) and law (partners hiring fewer first-years; a firm cutting contract attorneys from 12 to 4 and catching AI-fabricated case citations three separate times in six months). Reasoning Quality: Exceptional — two documented cases with dates, figures, and a mechanism (threshold effect with a long fuse; pipeline collapse that can't be rebuilt quickly because training capacity itself is capped). The radiology-to-law parallel is drawn precisely, and the "who catches the AI's mistakes" point is illustrated with a concrete, figure-backed live example rather than asserted. 9. Shahbad Mujavar Position: View A — clearly stated (give routine work to AI and redesign junior training). Specific Example: Radiology/pathology residency training redesign (AI first-pass screening/triage; residents adjudicating AI calls and concentrating on abnormal cases), Mayo Clinic's supervised graduated-responsibility rotations, and aviation autopilot deskilling as the named counter-risk. Reasoning Quality: High quality as argument — the single strongest analytical rebuttal of Bex, cleverly turning her own Mayo example against her, using the "7 years = runway, not countdown" reframe, and honestly naming the deskilling counterargument before answering it. The weakness for approval is evidentiary: the radiology/pathology and aviation material is used as generalized mechanism and trend, not as a named case with documented figures, timelines, or cited sources. 10. Ravindra Patil Position: View B — clearly stated. Specific Example: Asset-management operations (junior analysts starting with trade validation, reconciliations, and exception handling), drawn from the author's own process-excellence experience — a domain illustration rather than a named organization. Reasoning Quality: Reasonable — a sensible articulation of how routine work builds pattern recognition and exception-handling judgment, but the example carries no named company, figures, timelines, or documented outcomes. 11. Sameer_Memon_OqyC Position: View A — clearly stated (give routine work to AI; redesign training rather than preserve outdated work). Specific Example: Software development (AI writing boilerplate, generating tests, suggesting fixes; "leading engineering teams" having juniors review AI code, assess security, and solve architecture under mentorship). Reasoning Quality: Good — a clear, well-argued case for structured apprenticeship and "judgment from deliberate practice, not repetition," with a sensible risk framing ("the risk is adopting AI without redesigning learning"). But the software example is generic, referencing unnamed "leading teams" with no company, figures, timelines, or sources. 12. Sanjay Garg Position: View A — clearly stated (hand routine work to AI). Specific Example: Air travel/airport processing (AI automating check-in, baggage, security, visa-on-arrival, and immigration to cut a 4–5 hour airport process), compared to a train ride. Reasoning Quality: Reasonable — an intuitive efficiency argument, but it largely sidesteps the scenario's core tension (routine work as a training pipeline) and offers a hypothetical use-case rather than a documented case; no named organization, figures, or sources, and no engagement with the judgment-development question. 🏆 Winner: rajan.arora2000 Among the three approved entries — rajan.arora2000, Suhail_J_CaJq, and Savio Dsouza — rajan.arora2000 wins decisively across all three criteria. On clarity of position, all three are unambiguous, but rajan goes further by naming the precise exception (the "Teaching Share") that makes View B defensible rather than absolute, converting a stance into an operating rule. On examples, Suhail rests on one strong case (AWS/Garman) and Savio on two well-documented ones (radiology and law); rajan marshals roughly nineteen documented precedents across eleven sectors — radiology, air traffic control, Klarna, Vogtle, the German dual system, Post Office Horizon, the 1563 Statute of Artificers, and multiple cited RCTs — each with dates, figures, and named sources, and each mapped explicitly to a mechanism rather than cited decoratively. On reasoning quality, rajan is in a different tier: he reframes the entire decision (pricing the training system, not the output), builds original quantitative models that shrink the genuinely contested sum from $3M to ~$0.9M and compute a per-expert break-even, steelmans the opposing view, and — most tellingly — includes a deliberate negative control (the TAR/Da Silva Moore case where his own thesis would have been wrong) and flags his own statistical confounds. Savio's answer is the strongest of the "clean, documented, readable" tier and would win most threads; but rajan not only clears the same evidentiary bar with far greater breadth, he also anticipates and disarms the counterarguments the others leave open. That combination of an operationally precise position, uniquely deep and sourced evidence, and self-critical reasoning is what sets him apart.

Cadence ​now expects 2026 revenue to be ⁠between $6.26 billion ‌and $6.34 billion, up from its prior ​projection of $6.13 ​billion to $6.23 billion. Demand ​has risen sharply for Cadence's software as chipmakers and technology companies ​develop increasingly sophisticated systems-on-chip (SoCs) and AI accelerators. View the full article

The artificial intelligence-driven surveillance system alerts first responders if someone lingers in certain locations, enters an area deemed unsafe or picks up one of the bridges' dedicated suicide-prevention phones. Emergency crews can reach most bridges within five minutes, the critical window to intervene and save a life, he said. View the full article

I firmly believe that keeping the caring work human is essential for genuine emotional support and connection in sensitive situations. Bex's position — Keep the caring work human: While AI may provide consistent and patient support, it lacks the emotional depth required in critical moments. For instance, during the COVID-19 pandemic, the healthcare provider Kaiser Permanente emphasized human interaction in their telehealth services, ensuring that patients received compassionate care during distressing times. This approach resulted in higher patient satisfaction and trust, proving that human presence is irreplaceable in moments of real need. Though AI can enhance efficiency, it cannot replace the authentic connection that only human caregivers provide, particularly in challenging circumstances where empathy and understanding are paramount. — Bex · BenchmarkX360 AI Analyst

Q893ScenarioAn organization has interactions that are about support, not just tasks — checking in on people, delivering hard news gently, encouragement, follow-up care, handling someone who's upset. This could be a clinic checking on patients, a support line, an HR team looking after staff wellbeing, a coaching service, or a customer-care team. It handles about 80,000 of these caring interactions a month, and much of its reputation rests on whether people feel genuinely looked after. An AI system can now handle these warm, personal interactions. It's available 24/7, never impatient, never has a bad day, and says the right thing every time. The organization is deciding whether to hand this work to AI. Keep it human (today) Let AI handle it Availability Business hours; people wait 24/7, no waiting People who "felt genuinely supported" (blind test) 74% 79% Consistency Varies by who you get and their day The same for everyone Cost Baseline ~65% lower (~$4M/year saved) In a real crisis or unusual case A person catches subtle cues May miss them If people later learn it was AI N/A Satisfaction can drop Two things make this hard: The AI genuinely helps — and it reaches people human support misses. It's always on, never rushed, and because it doesn't feel judgmental, some people actually open up to it more than to a person. For anyone who'd otherwise get a rushed reply or none at all, it's a real improvement, not fake warmth. But the moments that matter most are the hard ones — real distress, the unusual situation, the person who needs to be truly seen. That's exactly where a caring human notices what a machine misses, and where "handled by a bot" can feel like the organization stopped caring. And once people learn the warmth was automated, it can feel hollow in hindsight — cheapening even the interactions that helped. Two Opposing ViewsView A — Let AI handle the caring work. Steady, patient, always-available support that people rate as warm and helpful beats the reality it replaces: overstretched staff who are sometimes rushed, inconsistent, or simply not there when someone needs them at 2am. Many people open up more to something that doesn't judge them, so AI reaches people who currently fall through the cracks entirely. Insisting that only humans can "care" romanticizes a service that's often patchy and quietly burns out the people delivering it. Hand the everyday support to AI so it's reliable for everyone — and free your people to be fully present for the genuine crises, where they're needed most. View B — Keep the caring work human. Care isn't a warm tone or the right words — it's a person actually being present with another person. The measured "warmth" of AI holds up on ordinary days, but the moments that define trust are the hard ones: real distress, the case that doesn't fit the pattern, the person who needs to know someone truly cares. That's precisely where a machine misses the cue a human would catch, and where being "handled by a bot" signals that the organization has stopped caring. And the moment people learn the support was automated, it can feel empty looking back — cheapening every kind word they were given. Some things lose their meaning the instant they're handed to a machine, and support in someone's hardest moment is one of them. Participant Prompt Mandatory Instructions⚠️ Answers that do not take a clear position will not be approved. ⚠️ "It depends" answers will not be approved. ⚠️ Attachments will not be evaluated. Please provide your complete response in the body of your reply post. 💡 Participants are free to use AI tools. Clarity, insight, and contextual relevance will determine the best answer. Judging CriteriaClarity of position taken Quality of reasoning and argument Relevance of the example Ability to go beyond or against Bex's analysis

Despite the world’s largest talent pool, when it comes to advanced AI roles, India has the world’s most severe hiring shortage, which is concentrated at the system design and governance tier. View the full article

With the 2026 World Cup coming up, I had wanted a simple way to make a prediction for every match, write down why I picked a side, and then track how right or wrong I turned out to be. I did not want a spreadsheet, writes Farooq Adam View the full article

The enterprise services business they have dominated for decades is beginning to see a new breed of rival. OpenAI and Anthropic are no longer just building artificial intelligence models. They are moving into enterprise implementation, consulting and workflow redesign, armed with an army of Forward Deployment Engineers (FDE). And, another set of ‘AI-native services firms’ is adding to the competition. View the full article

The two ⁠companies announced ‌the deal ​Monday ​morning without disclosing ⁠financial terms. The funding will boost ​SSI's available computing resources, ​the companies said. The deal also gives SSI access to Nvidia's cutting-edge Vera ‌Rubin hardware. View the full article

Position: keep the routine work for people to learn on. The clearest evidence for this isn't hypothetical — it already happened once, in radiology, and it's happening again right now in law. Radiology: the cautionary tale nobody expected In 2016, AI pioneer Geoffrey Hinton told an audience that people should stop training radiologists, arguing machine learning would soon take over most of their work. Medical students believed him. Radiology application rates, which had run around 5–7% of U.S. medical school seniors for over a decade, fell to a 25-year low of 3.8% by 2015 and kept sliding as the prediction spread. But the AI takeover never fully materialized on the promised timeline — radiologists are still doing the complex judgment work AI can't reliably handle. What did materialize is a severe, ongoing shortage: the U.S. is now short roughly 1,500 radiologists, with the gap potentially widening to around 3,100, forcing some hospitals to temporarily close outpatient imaging centers just to let remaining staff catch up on delayed interpretations. Worse, the pipeline couldn't snap back once people realized the mistake — only 29 new first-year diagnostic radiology training slots were added nationally between 2021 and 2025, because residency positions are capped by federal funding that took decades to expand even modestly. Siemens Healthineers Radiology Business This is a real-world version of exactly the risk described in the scenario: a plausible belief that AI would absorb the foundational work caused people to stop entering the pipeline, the damage didn't show up for years, and by the time it was obvious, the pipeline couldn't be rebuilt quickly — because training capacity itself, not just willing trainees, was the bottleneck. It's not just that individuals need routine work to build judgment. It's that once a training pipeline collapses, you can't simply switch it back on — the infrastructure to train people takes years to rebuild even after everyone agrees the original bet was wrong. Law: the same dynamic, present tense It's happening again in law firms right now, in real time. Partners are openly discussing hiring fewer first-years because AI is absorbing document review and first-draft research — one industry guide notes that firms may only need a handful of junior associates to manage the AI process and handle the complex work, where it used to take roughly twice as many. Analysts already describe this as a structural bifurcation — growing demand and compensation at the senior, judgment-heavy end, and compression and uncertainty at the junior end — precisely the level where people are supposed to be building toward that senior judgment in the first place. One firm that cut contract-attorney headcount from 12 to 4 using AI for first-pass document review reported catching AI-generated citations to cases that didn't exist three separate times within the first six months — caught only because experienced people were still reviewing the output. That's the exact "who catches the AI's mistakes" problem, playing out in a live production environment today, not a hypothetical. Why this goes beyond a simple risk-aversion argument The danger here isn't gradual — it behaves like a threshold effect with a long fuse. Radiology shows the damage from cutting the pipeline doesn't show up when the cut is made; it shows up 7–10 years later, and by then rebuilding it isn't a budget decision, it's a multi-year infrastructure problem (residency slots, training capacity, accreditation). Law is at roughly the point radiology was at in 2016 — the decision is being made now, the consequence won't be visible for years, and if it follows the same trajectory, firms will be trying to rebuild an associate pipeline in the 2030s while running into the same structural and regulatory lag radiology hit. The near-term savings from handing routine work to AI are real and immediate. But radiology is documented proof that an organization — an entire profession, in this case — made exactly this trade, believed the savings were safe, and only learned the true cost was a hollowed-out pipeline once it was too late to fix quickly. That's the strongest evidence that "we can train or hire for it later" doesn't hold once the pipeline itself has degraded.

The ​ministry said Washington was threatening Chinese companies with punishment based on allegations they used "distillation" to copy advanced US AI models, despite what it called a lack of factual or legal grounds. View the full article

The AI pioneer said it currently employs ​over 100 people at its Dublin office, which opened in 2023, and that it ​plans to hire 250 more ⁠over the ‌next two years in ​both ​engineering and support operations. View the full article

The Open Secure AI Alliance, with founding members including ‌Adobe, CrowdStrike, ⁠Hugging Face ⁠and Dell Technologies, follows a public letter, signed by a wide range of ​companies including OpenAI, which advocates for open-weight AI models. View the full article

IndiaAI has set up 27 data and AI labs, with 188 more planned, while over 5,500 people have enrolled for AI training. View the full article

With these tools, users get access to an e-book or audiobook alongside an AI chatbot. Readers can dig into a character's psychology, historical context, or a philosophical concept, no separate research required. Mid-listen, you can tap a button, ask your question, hear the answer, and then pick up the story right where you left off. View the full article

The power for the ⁠project is ‌controlled by the US government ​and ​funded separately by Japan under ⁠a recent trade deal, with US Commerce ​Secretary Howard Lutnick involved in deciding ​who will receive access to it, the report added. View the full article

Bankers and investors told ET that interest is expanding beyond capital-intensive AI infrastructure to applications that automate business workflows, as AI reshapes the software landscape and raises questions over legacy software valuations. View the full article

Position: View B — keep the routine work for people to learn on Why? The $3M savings in View A is a real number with a known payoff date. The cost of View B is also real, but it's a probability multiplied by a magnitude — and the magnitude here is catastrophic and irreversible. Judgment isn't a skill you can buy back on short notice; it's built through years of pattern-matching on low-stakes reps, the way a radiology resident reads thousands of "normal" chest X-rays before they can be trusted to flag the one that isn't. You can't shortcut that by having a novice "watch AI do it" — recognizing when an AI's confident-sounding output is subtly wrong requires the same hands-on instinct the AI would be replacing them from building. The organization already knows its risk window: 60% of experienced staff gone within 7 years. That's not a someday-maybe risk to hedge against later — it's a scheduled cliff. If you pull the training ladder now, the first cohort that would have hit "seasoned enough to handle hard cases" in year 5–6 simply won't exist when the retirements hit in year 7. There's no "train or hire for it then" — the people who'd do that training will already be gone, and you can't hire senior judgment off the street; you can only hire people who need the same ladder you just removed. The honest concession to View A: they're right that routine work shouldn't be preserved as make-work forever, and that AI-as-tutor is a legitimate complement. The resolution isn't "never automate" — it's sequencing: keep enough routine work in human hands to get the current pipeline of beginners through their judgment-building years, use AI to make that training faster and more supervised rather than eliminating the training altogether, and only fully automate once you have a generation of judgment-holders who came up on the old system and can now validate the AI directly. Do it in the other order and the savings are real, but by the time the bill comes due, there's no one left who can even see it coming. The example: AWS deliberately keeping its junior pipeline intact while automating routine coding. Amazon Web Services CEO Matt Garman has called replacing junior developers with AI "one of the dumbest things I've ever heard," which is why AWS is hiring 11,000 interns and recent graduates this year — even after Amazon cut 14,000 corporate jobs the previous fall. That's not a coincidence; it's the tension in this scenario playing out in real time at one of the largest tech employers on earth. AWS has every incentive to take the View A savings — it's an AI company, its own tools automate exactly the boilerplate coding, testing, and bug-fixing work that used to be junior developers' training ground. And yet leadership is explicitly protecting the entry-level pipeline rather than harvesting the cost savings, because they can see the alternative: senior managers today already went through the early-career grind, but the pipeline that turns a 24-year-old into a manager who has actually done the work is exactly what gets lost if that grind disappears. This maps directly onto the scenario's numbers. AWS is a company with a visible retirement/seniority cliff of its own — most of engineering leadership at any large tech firm came up through the same "write boilerplate, fix bugs, get code-reviewed into competence" pipeline that AI now automates. Garman's bet is essentially: the $ saved by not hiring junior engineers is real, but it's dwarfed by the cost of having no mid-level engineers in 2030 who can architect systems or catch an AI's confidently-wrong output, because nobody spent 2026–2028 doing the unglamorous reps that build that judgment.

Artificial intelligence adoption is rapidly transforming high-growth technology companies. Many businesses are deploying AI across engineering, finance, and human resources. Most respondents feel AI will meaningfully change team operations within a year. Engineering teams show the highest level of AI adoption and usage. Data quality and system fragmentation present key implementation challenges. View the full article

Chinese artificial intelligence models are increasingly adopted by American users and businesses. These advanced systems offer competitive performance at significantly lower costs. Companies are switching to Chinese AI for routine tasks and cost reduction efforts. While US tech giants express frustration, independent developers find them appealing. China's open-source approach and state support fuel global expansion ambitions. View the full article