How can we reliably take a look at whether or not giant language fashions truly perceive Indian languages and tradition in actual world contexts? OpenAI has launched IndQA, a benchmark that evaluates how effectively AI fashions perceive and cause about questions that matter in Indian languages throughout cultural domains. Why IndQA? OpenAI states that about 80 p.c of individuals worldwide don’t converse English as their major language. But most benchmarks that measure non English capabilities are nonetheless slim and infrequently depend on translation or a number of selection codecs. Benchmarks equivalent to MMMLU and MGSM at the moment are close…
Author: Naveed Ahmad
The Movement Image Affiliation (MPA) has despatched Meta a cease-and-desist letter demanding that it cease utilizing the time period “PG-13,” as first reported by The Wall Street Journal. Final month, Meta introduced that teenagers on Instagram would, by default, solely see content material that adheres to PG-13 film scores. Two weeks later, the MPA despatched Meta a cease-and-desist letter, asserting that Meta’s declare that content material on teen Instagram accounts would observe PG-13 pointers is “actually false and extremely deceptive.” The corporate states that its movie-rating system can’t be in comparison with Meta’s content material restrictions, which it says “seem…
How can consistency coaching assist language fashions resist sycophantic prompts and jailbreak model assaults whereas preserving their capabilities intact? Massive language fashions typically reply safely on a plain immediate, then change habits when the identical activity is wrapped with flattery or function play. DeepMind researchers suggest constant coaching in a easy coaching lens for this brittleness, deal with it as an invariance downside and implement the identical habits when irrelevant immediate textual content modifications. The analysis workforce research two concrete strategies, Bias augmented Consistency Coaching and Activation Consistency Coaching, and evaluates them on Gemma 2, Gemma 3, and Gemini 2.5…
Epic Video games CEO Tim Sweeney is calling Google’s proposal in its antitrust settlement with the Fortnite maker a “complete resolution” that “genuinely doubles down” on Android’s imaginative and prescient of being an open platform. The businesses on Tuesday reached a settlement that sees the search large agreeing to Android app retailer reforms that embody decreasing charges and enabling extra competitors. Beneath the brand new proposal, which nonetheless requires the choose’s approval, Google will permit Android app builders to level customers to various cost mechanisms inside their apps and thru exterior net hyperlinks. It additionally caps the charges Google is…
A raft of voice-based {hardware} gadgets have emerged, geared toward companionship, productiveness, or private progress. These embrace card-shaped gadgets from Plaud and Pocket; pendants from Buddy, Limitless, and Taya; and a wristband from Bee, which is now a part of Amazon. Now, two former Meta staff who labored on interface design have launched Sandbar, a startup that has created a hoop referred to as Stream for related functions. The corporate calls the ring “a mouse for voice” as a result of it might take notes, make it easier to work together with an AI assistant, and likewise allow you to…
Cybersecurity is an enormous sector, however startups within the class usually tend to be acquired than go public. Even Wiz, which for a time held the title of the fastest-growing startup, deserted its IPO ambitions when it agreed to promote to Google earlier this 12 months. Previously few years, there have been scant few vital cybersecurity listings: SentinelOne IPO’d in 2021, Rubrik did so final 12 months, and Netscope went public in September. Armis, a nine-year-old cybersecurity startup based mostly out of San Francisco, intends to observe in these corporations’ footsteps. The corporate mentioned on Wednesday that it has raised a $435 million pre-IPO spherical…
Everybody has an inside monologue. If you’re commuting on the prepare, driving a motorcycle, or within the bathe, likelihood is you are desirous about the day forward, duties you should do, or possibly simply mulling over a dialog you had the night time earlier than. A lot of this stays in our brains, quickly to be forgotten or pushed away when the prepare involves the station. However what if you happen to may have all of it subtly recorded in a single place, prepared so that you can digest afterward?That is what a brand new firm referred to as Sandbar…
E-commerce software program supplier Shopify is bullish on AI-powered purchasing brokers, citing AI as an “unbelievable device” to allow extra entrepreneurs and calling it the “greatest shift in know-how for the reason that web” throughout its third-quarter earnings name. The corporate, which partnered with ChatGPT maker OpenAI in September, reported that visitors from AI instruments to its on-line shops is up 7x since January of this yr, and purchases attributed to AI-powered search have elevated by 11x. Based on Shopify president Harley Finkelstein, the corporate’s benefit within the AI period comes from its capability to entry the info from tens…
Bild från: PewDiePieFelix Kjellberg (PewDiePie) publicerade en video där han visade sitt hemmabyggda AI‑projekt ChatOS.10‑GPU mini‑datacenter med PCIe‑bifurcation och blandade RTX 4000 Ada + moddade RTX 4090, byggt för lokal LLM‑körning.Kör öppna modeller som Llama 70B och Qwen i ett eget UI (“ChatOS”) med verktyg som webbsök, RAG, röst och minne.Riggen har även donerat beräkning till forskning via Folding@home när den varit ledig.PewDiePie har byggt ett självhostat AI-system kallat ChatOS på en lokal, multi‑GPU‑rack för att köra stora öppna modeller utan molntjänster. Det är ett coolt rätt imponerande AI-projekt eller mini‑lab av svensken Felix Kjellberg (PewDiePie).ChatOS är ett egenbyggt webbgränssnitt…
Simply over three years after taking the reins as the highest chief of Sequoia Capital, Roelof Botha is stepping down as senior steward of the storied VC agency. The agency introduced Tuesday that companions Alfred Lin and Pat Grady will succeed him as co-stewards. Lin joined the storied agency in 2010, the place he has led main investments into category-defining corporations like Airbnb, DoorDash, and Kalshi. In the meantime, Pat Grady has been a accomplice for almost 19 years and has led Sequoia’s growth-stage investing since 2015, backing iconic corporations reminiscent of ServiceNow, OpenAI, and the authorized AI platform Harvey.…