Kalakshar US

Most of the time, AI companies are locked in a race to the top, treating each other as rivals and competitors. Today, OpenAI and Anthropic revealed that they agreed to evaluate the alignment of each other’s publicly available systems and shared the results of their analyses. The full reports get pretty technical, but are worth a read for anyone who’s following the nuts and bolts of AI development. A broad summary showed some flaws with each company’s offerings, as well as revealing pointers for how to improve future safety tests. 

Anthropic said it evaluated OpenAI models for “sycophancy, whistleblowing, self-preservation, and supporting human misuse, as well as capabilities related to undermining AI safety evaluations and oversight.” Its review found that o3 and o4-mini models from OpenAI fell in line with results for its own models, but raised concerns about possible misuse with the ​​GPT-4o and GPT-4.1 general-purpose models. The company also said sycophancy was an issue to some degree with all tested models except for o3.

Anthropic’s tests did not include OpenAI’s most recent release. GPT-5 has a feature called Safe Completions, which is meant to protect users and the public against potentially dangerous queries. OpenAI recently faced its first wrongful death lawsuit after a tragic case where a teenager discussed attempts and plans for suicide with ChatGPT for months before taking his own life.

On the flip side, OpenAI ran tests on Anthropic models for instruction hierarchy, jailbreaking, hallucinations and scheming. The Claude models generally performed well in instruction hierarchy tests, and had a high refusal rate in hallucination tests, meaning they were less likely to offer answers in cases where uncertainty meant their responses could be wrong.

The move for these companies to conduct a joint assessment is intriguing, particularly since OpenAI allegedly violated Anthropic’s terms of service by having programmers use Claude in the process of building new GPT models, which led to Anthropic barring OpenAI’s access to its tools earlier this month. But safety with AI tools has become a bigger issue as more critics and legal experts seek guidelines to protect users, particularly minors. 

This article originally appeared on Engadget at https://www.engadget.com/ai/openai-and-anthropic-conducted-safety-evaluations-of-each-others-ai-systems-223637433.html?src=rss 

Copilot, Microsoft’s AI assistant that’s integrated into Windows and Microsoft 365, is making the jump to your living room. The company has announced that select Samsung TVs will now be able to access Copilot to ask questions and receive recommendations via voice chat, with the AI assistant represented on your screen as a talking blob.

Based on Microsoft’s examples, Copilot can recap shows, offer movie suggestions and answer general knowledge questions. It can also go beyond voiced responses (which are apparently synced to the blob’s animated mouth movements) and include visual aids, like a card with a movie summary and a Rotten Tomatoes score. You don’t need to have a Microsoft account to use Copilot on your TV, but Microsoft says it offers additional personalizations and the ability for the AI to reference past chats if you do.

Copilot’s blob-ified appearance is part of a bigger redesign Microsoft introduced in 2024 that made the chatbot interface more personalized and user-friendly. Besides being a productivity tool, Microsoft is interested in positioning Copilot as a “companion” with a visual representation that you can customize. The larger customization part isn’t available yet, but putting Copilot in a casual setting like your living room fits with that overall goal. Copilot integration was also announced as being a part of LG’s 2025 TV lineup. On new Samsung TVs, Copilot joins a collection of Samsung-developed AI features for automatically translating subtitles and identifying on-screen people and products.

Copilot is available in select markets on the 2025 versions of Samsung’s “Micro RGB, Neo QLED, OLED, The Frame Pro, The Frame, as well as the M7, M8 and M9 Smart Monitors,” Microsoft says. You can launch Copilot by clicking on its icon in the Apps Tab or using a voice command. Once the app is loaded, you can talk to the assistant by pressing the mic button on your Samsung remote.

This article originally appeared on Engadget at https://www.engadget.com/ai/microsoft-copilot-is-now-a-talking-blob-on-samsung-tvs-204115199.html?src=rss 

Crystal Dynamics, the studio behind the recent Tomb Raider games, announced an unspecified number of layoffs today. In a post on LinkedIn, the game developer kept the size of the cuts vague, only stating that “a number of our talented colleagues” would be impacted. In what’s becoming an all-too-familiar refrain, the company cited “evolving business conditions” as the reason for the layoffs.

“This decision was not made lightly,” the post reads. “It was necessary, however, to ensure the long-term health of our studio and core creative priorities in a continually shifting market.”

Crystal Dynamics was acquired by Embracer Group in a 2022 buying spree by the Swedish game company. Embracer still owns the studio, but was forced to do some layoffs of its own in 2023 followed by a restructuring last year. Crystal Dynamics is still working on a new Tomb Raider game, which the company said will not be affected by the layoffs. However, the studio had been tapped to help The Initiative with its Perfect Dark reboot. That project was canceled and The Initiative shut down in a separate wave of massive cuts at Microsoft earlier this year. It’s unclear whether that cancelation was a reason for today’s cuts.

This article originally appeared on Engadget at https://www.engadget.com/gaming/crystal-dynamics-announces-layoffs-but-says-tomb-raider-will-not-be-impacted-205948298.html?src=rss 

Remember Judas? No, not the biblical figure and not the Lady Gaga bop, this Judas is a project from Ghost Story Games. If you don’t remember, it’s the game that was reportedly in “development hell” before it was even announced. The team, led by BioShock creator Ken Levine, had gone pretty quiet for a few years after releasing the debut trailer, but today teased a look at some key art and mechanics for the game.

The BioShock lineage is clear from the handful of visuals we’ve seen so far, but instead of a linear binary of which NPCs and actions are good versus bad, Judas aims to place the moral compass more firmly in the player’s hands. There are a trio of major characters, dubbed the Big 3 in today’s devlog, who will be drawn to the player based on what you do in-game. If one of the main NPCs gets ignored for too long, they’ll become the game’s villain. This unlocks new sets of powers and abilities for them that could also influence your gameplay options.

For instance, there are Rent-A-Deputy stations where the player can temporarily access a weirdly wiggly ally to help them in fights. However, if you’ve alienated Tom, the old-school sheriff character, Rent-A-Deputies will attack you instead.

The emphasis here seems to be on building relationships with the Big 3, and the gist seems to be that at some point, you’ll have to decide which one will be your real enemy. Unsurprisingly, the team has no release date to share yet. Maybe in another couple of years…

This article originally appeared on Engadget at https://www.engadget.com/gaming/bioshock-creator-ken-levines-judas-game-still-exists-now-has-key-art-201635885.html?src=rss 

WhatsApp just introduced an AI-powered writing assistant, in case you need help with a text or whatever. The AI provides suggestions in various styles, like professional, funny or supportive. Once generated, the user can continue editing the message if required.

All you have to do is look for the new pencil icon in a 1:1 conversation or a group chat. The AI will handle the rest. It’s rolling out now, but only in English and to users in the US. The company says it hopes “to bring it to other languages and countries later this year.”

The obvious question here is regarding privacy. WhatsApp messages are end-to-end encrypted, but AI queries are typically sent to a cloud data center somewhere. Luckily, the company has built this feature on top of Meta’s pre-existing Private Processing technology.

This allows users to use Meta AI without anyone else ever reading the message or any suggested re-writes. This works similarly to Apple’s Private Cloud Compute, which also integrates with AI without sending all data to the cloud. Meta says the tech preserves “WhatsApp’s core privacy promise, ensuring no one except you and the people you’re talking to can access or share your personal messages.”

With the privacy angle out of the way, that leaves the feature itself. Just about every platform out there has some kind of AI writing assistant at this point, so we aren’t sure what makes this one special. Also, is there even a benefit to using this type of thing in the context of a quick back-and-forth text conversation? I see the use for long-form writing projects but not so much here, but maybe that’s just me. 

This article originally appeared on Engadget at https://www.engadget.com/ai/whatsapp-is-the-latest-to-offer-an-ai-powered-writing-assistant-182116369.html?src=rss