Blogroll

True crime might be Netflix's most consistent genre in terms of hit rate. True-crime documentaries, including The Perfect Neighbor and Amy Bradley Is Missing, dominated the Netflix algorithm last year. The sentiment also includes scripted true crime, and Suitcase Killer: The Melanie McGuire Story is the latest hit for the streamer.

The rollout of ChatGPT age prediction is now underway, according to an announcement OpenAI made Tuesday.

Last fall, the company indicated it would introduce age prediction as a safety measure for teens. OpenAI has been sued by parents of teens who died by suicide after ChatGPT allegedly coached them to end their lives or didn't respond appropriately to their discussions of psychological distress. OpenAI has denied the allegations in the first of those lawsuits.

In December, the company introduced an update to its Model Spec, which guides how its AI models should behave. The update focused on principles for responding to under-18 users in high-stakes situations.

SEE ALSO: OpenAI says its mystery AI wearable is on track for 2026 as AI earbuds rumors spread

The details made public on Tuesday explain that age prediction will launch on consumer plans.

ChatGPT's age prediction model estimates a user's age based on behavior and signals from their account, like when the person is active during the day, long-term usage patterns, how long the account has existed, and the user's stated age.

If the user's age is assessed as under-18, their account should not be exposed to graphic violence, depictions of self-harm, and sexual, romantic, or violent role play, among other types of potentially harmful content.

Teens who tell OpenAI they're under 18 upon opening a ChatGPT account are automatically subject to such safeguards. If OpenAI isn't confident about a user's age, their account will default to safer settings.

OpenAI said that adult users whose accounts are mistakenly placed in the under-18 experience can confirm their age by submitting a selfie to Persona, a third-party identity verification service. OpenAI did not provide additional information about how ID documents would be retained. In Oct. 2025, a third-party vendor used by the messaging platform Discord was breached, exposing upwards of 70,000 government IDs.

OpenAI said it planned to improve age-prediction accuracy and make further improvements based on observations from the initial rollout.

A FlexJobs survey in August 2025 found that even though most employees have paid time off, almost one-quarter didn’t take a single vacation day. Meanwhile, our inboxes can often feel like a mess, as the average person receives over 100 e-mails a day.

Only a couple weeks removed from seeing a bunch of weird robots at CES, scientists have unveiled a new innovation in the world of creepy robotics — the detachable crawling robotic hand.

Of course, one person's creepy is another person's awesome, and it's an impressive little bot that invites an instant comparison to Thing, the sentient, crawling hand from The Addams Family.

As published in Nature Communications, a group of scientists, and first author Xiao Gao, recently unveiled the new type of robotic hand. Videos published alongside the paper show how the hand can detach itself from its arm and use its fingers to crawl around like a spider.

The detachable robot isn't constrained by the limitations of the human hand. While its six fingers aren't too far off from the five digits on our own hands, everything else about this little creepy crawler is pretty divergent. For instance, the fingers can bend backwards as well as they bend forwards, so the robot can hold objects against both sides of its base (or palm, if you want to map it to human anatomy) at the same time. It's even capable of carrying small objects on its back (or belly) while it does so, using its free fingers for locomotion.

Obviously, these features would be most useful in industrial settings, but maybe someday robots like this will find their way into regular people's homes, too.

2026's television slate has no shortage of familiar faces returning to our screens. Not only will series such as Avatar: The Last Airbender and The Boys take us back to fantastical realms, but grounded series like The Pitt will return to shine a light on the struggles that the real heroes in our world face.

If you’re after a car that won’t leave you stranded and won’t guzzle gas, this Toyota hybrid has you covered.

Fast charging is an equally complex and fascinating technology that allows phones to draw much higher levels of power for quicker battery top-ups. What many people don’t realize is that it relies on constant communication between the phone and the charger, and when that communication breaks down, charging falls back to much slower speeds.

Audiobooks are a great way to “read” when you’re not able to read for real. However, if you switch between audio and text, you know how annoying it can be to sync your progress—especially across different mediums. Spotify is working on a clever solution.

What started as a compromise eventually led to a new preference for me. I had always thought I needed 18V tools for DIY projects around the house, but I was sorely mistaken. Here’s why my Milwaukee M12 tools are more than enough for DIY projects around the house, and how they became my go-to tools for everything.

A developer has submitted draft code to the KDE team that could turn the Plasma desktop environment into a fully functional 3D VR interface. This is a huge move for Linux desktop users who want to ditch flat screens and jump into immersive computing.

TerraMaster is best known for its network attached storage (NAS) devices, but the company also makes some external storage enclosures. TerraMaster has now revealed the D1 SSD Pro, which turns any NVMe SSD into a super-fast Thunderbolt 5 external drive.

The end of Yellowstone left a gaping hole in the hearts of viewers everywhere. As that hole has grown, so, too, has the search for the perfect series to fill the void. Luckily, I’ve got that answer—and no, it’s not Longmire. Instead, we're going completely unorthodox and trading horses for Harleys and cowboys for outlaw bikers.

TL;DR: Live stream Tottenham vs. Borussia Dortmund in the Champions League for free on Virgin Media Player. Access this free streaming platform from anywhere in the world with ExpressVPN.

The Champions League knockout rounds are fast approaching. Tottenham face off against Borussia Dortmund in the seventh round of group stage games, with both teams still very much in the qualification picture. Both sides are sat on 11 points after six games, and could jump into the top eight with three points. The prospect of avoiding an extra two games in the playoffs should be plenty of motivation here.

Dortmund face Inter in their final group stage game. Tottenham take on Eintracht Frankfurt. Neither final group stage game is going to be easy, so this matchup takes on greater significance in the race to avoid the playoffs.

If you want to watch Tottenham vs. Borussia Dortmund in the Champions League from anywhere in the world, we have all the information you need.

When is Tottenham vs. Borussia Dortmund?

Tottenham vs. Borussia Dortmund in the Champions League kicks off at 8 p.m. GMT on Jan. 20. This fixture takes place at the Tottenham Hotspur Stadium.

How to watch Tottenham vs. Borussia Dortmund for free

Tottenham vs. Borussia Dortmund is available to live stream for free on Virgin Media Player.

Virgin Media Player is geo-restricted to Ireland, but anyone can access this free streaming platform with a VPN. These tools can hide your real IP address (digital location) and connect you to a secure server in Ireland, meaning you can unblock Virgin Media Player to stream the Champions League for free from anywhere in the world.

Live stream Tottenham vs. Borussia Dortmund for free by following these simple steps:

1. Subscribe to a streaming-friendly VPN (like ExpressVPN)

2. Download the app to your device of choice (the best VPNs have apps for Windows, Mac, iOS, Android, Linux, and more)

3. Open up the app and connect to a server in Ireland

4. Visit Virgin Media Player

5. Watch Tottenham vs. Borussia Dortmund for free from anywhere in the world

Opens in a new window Credit: ExpressVPN ExpressVPN (1-Month Plan) $12.95 only at ExpressVPN (with money-back guarantee) Get Deal

The best VPNs for streaming are not free, but most do offer free-trials or money-back guarantees. By leveraging these offers, you can access free live streams of the Champions League without actually spending anything. This obviously isn't a long-term solution, but it does give you enough time to stream Tottenham vs. Borussia Dortmund (plus more Champions League fixtures) before recovering your investment.

What is the best VPN for Virgin Media Player?

ExpressVPN is the best choice for bypassing geo-restrictions to stream live sport on Virgin Media Player, for a number of reasons:

• Servers in 105 countries including Ireland

• Easy-to-use app available on all major devices including iPhone, Android, Windows, Mac, and more

• Strict no-logging policy so your data is secure

• Fast connection speeds free from throttling

• Up to 10 simultaneous connections

• 30-day money-back guarantee

A one-year subscription to ExpressVPN is on sale for $99.95 and includes an extra three months for free — 49% off for a limited time. This plan includes a year of free unlimited cloud backup and a generous 30-day money-back guarantee. Alternatively, you can get a one-month plan for just $12.95 (with money-back guarantee).

Live stream Tottenham vs. Borussia Dortmund in the Champions League for free with ExpressVPN.

TL;DR: Live stream Inter Milan vs. Arsenal in the Champions League for free with a 30-day trial of Amazon Prime. Access this free live stream from anywhere in the world with ExpressVPN.

The Champions League is back with another batch of huge fixtures involving Europe's elite. Inter Milan take on Arsenal in one of the best games of the round. Arsenal have won every group stage game so far and sit on top of the standings. Inter aren't too far back, and will be looking to safeguard their place in the top eight with a victory at home.

Inter will take on Dortmund in their final group stage game next week. That's considerably tougher than Arsenal's final game against Kairat. Will Arsenal rest players in this matchup? With a stacked schedule over the next few weeks, it's a strong possibility.

If you want to watch Inter Milan vs. Arsenal in the Champions League for free from anywhere in the world, we have all the information you need.

When is Inter Milan vs. Arsenal?

Inter Milan vs. Arsenal in the Champions League kicks off at 8 p.m. GMT on Jan. 20. This fixture takes place at the San Siro.

How to watch Inter Milan vs. Arsenal for free

Inter Milan vs. Arsenal in the Champions League is available to live stream on Prime Video in the UK, but you don't need to be subscribed to Amazon Prime to watch this fixture. Instead, you can watch Inter Milan vs. Arsenal for free with a 30-day trial of Amazon Prime.

If you're abroad for this fixture, you might need to use a VPN to watch the Champions League for free on Prime Video. This process is straightforward:

1. Sign up for a 30-day Amazon Prime trial (if you're not already a member)

2. Subscribe to a streaming-friendly VPN (like ExpressVPN)

3. Download the app to your device of choice (the best VPNs have apps for Windows, Mac, iOS, Android, Linux, and more)

4. Open up the app and connect to a server in the UK

5. Live stream Inter Milan vs. Arsenal for free on Prime Video

Opens in a new window Credit: ExpressVPN ExpressVPN (1-Month Plan) $12.95 only at ExpressVPN (with money-back guarantee) Get Deal

The best VPNs for streaming are not free, but most do offer free-trials or money-back guarantees. By leveraging these offers, you can watch Inter Milan vs. Arsenal on Prime Video without committing with your cash. This isn't a long-term solution, but it does give you enough time to stream select Champions League fixtures before recovering your investment.

What is the best VPN for Prime Video?

ExpressVPN is the best choice for bypassing geo-restrictions to stream live sport on Prime Video, for a number of reasons:

• Servers in 105 countries including the UK

• Easy-to-use app available on all major devices including iPhone, Android, Windows, Mac, and more

• Strict no-logging policy so your data is secure

• Fast connection speeds free from throttling

• Up to 10 simultaneous connections

• 30-day money-back guarantee

A one-year subscription to ExpressVPN is on sale for $99.95 and includes an extra three months for free — 49% off for a limited time. This plan also includes a year of free unlimited cloud backup and a generous 30-day money-back guarantee. Alternatively, you can get a one-month plan for $12.95 (with money-back guarantee).

Stream Inter Milan vs. Arsenal in the Champions League for free with ExpressVPN.

SAVE $1,500: The 55-inch Samsung Odyssey Ark 4K curved gaming monitor is on sale at Woot for $1,199.99, down from the list price of $2,699.99. That's a 56% discount and better sale price than we've ever seen at Amazon.

Opens in a new window Credit: Samsung 55-inch Samsung Odyssey Ark 4K curved gaming monitor $1,199.99 at Woot
$2,699.99 Save $1,500.00   Get Deal

Anyone who spends a decent amount of time gaming understands the importance of a solid gaming monitor. Some models can actually be the death of you, both in terms of frustration from a poor display and lag that causes your character to die in-game. If you're in the mood for a major "treat yourself" moment and you want a first-class gaming monitor, check out this limited-time deal.

As of Jan. 20, the 55-inch Samsung Odyssey Ark 4K curved gaming monitor is on sale for $1,199.99 at Woot, marked down from the list price of $2,699.99. That's a major 56% discount that takes $1,500 off the normal price. This sale beats the lowest we've ever seen at Amazon. Prime members get free shipping at Woot.

If you're looking for a mega sized gaming monitor with premium stats, this is your deal. Samsung consistently makes some of the best curved gaming monitors and today's deal on the giant 55-inch Ark 2nd Gen is a steal. The 1000R curve will bring a new experience to every game you play or show you stream on the monitor. With this size, you might not even want to use your TV anymore.

SEE ALSO: The giant 49-inch Samsung Odyssey OLED G9 curved gaming monitor is within $2 of a record-low price at Amazon

Of course, you're in line for ideal gaming stats like a 165Hz refresh rate and 1ms response time. Adjust into cockpit mode to take off on epic gaming adventures this winter. Samsung also equipped this monitor with four corner speakers and two central woofers.

Keep in mind Woot deals tend to sell out in an instant, so hop on this one if you're in the mood for a major treat. Amazon is selling this model at its full price of $2,699.99, so snagging this Woot deal is the way to go. Plus, you'll get free shipping if you sign into your Amazon Prime account at Woot.

Finding a performance sedan that’s thrilling on the road and dependable enough to handle hundreds of thousands of miles is rare, but one model manages both with surprising ease. While many sporty four-doors sacrifice reliability for speed or focus on short-term thrills, this performance-oriented sedan combines engaging driving dynamics with real-world durability you can count on year after year. For drivers who want the best of both worlds, excitement and long-term usability, it stands out in a crowded segment.

You should perhaps double-check the apps you use — especially if you're into AI.

A new project called Firehound from security firm CovertLabs tracks the apps that leak the most data, and the Top 10 was rife with AI apps. Here's a screenshot of the data from Firehound's website.

Lots of AI. Credit: Screenshot: Firehound

As you can see, the Top 10 list from Firehound has lots of AI apps. The project claims all kinds of data, from email addresses, to chat history, to names were left accessible by apps. One would hope that apps would work to fix such vulnerabilities moving forward, should they exist.

This Tweet is currently unavailable. It might be loading or has been removed.

Moving forward, it's probably best for users to tread carefully when downloading an app. It's worth stopping to think if it's a trustworthy product. And, of course, it's probably worth being cautious when sharing personal or professional information with an AI chatbot. You never know if it'll become public.

The best Pokémon TCG deals this week, at a glance: Best Pokémon TCG Elite Trainer Box Deal Pokémon TCG Phantasmal Flames Elite Trainer Box $79.94 (Save $26.95) Get Deal Best Pokémon TCG Booster Bundle Deal Pokémon TCG: Mega Evolution Phantasmal Flames Booster Bundle $52.70 (Save $12.29) Get Deal Best Pokémon TCG: Collector Chest Deal Pokémon TCG: Collector Chest Fall 2025 $37.99 (Save $20) Get Deal

Whether you're a longtime Pokémon TCG player and collector or just getting into the game, it's always good to save some cash when you can. These cards are hotter than ever, years after their debut, with demand at an all-time high for sets that released years ago, let alone those with more recent debuts.

So when you can snag a deal that works in your favor and nets you a set of awesome cards, it's a good idea to pounce on it, as is usually the case with most Pokémon-related things. With that in mind, however, how do you find some of the best deals on offer in the world of Pokémon TCG?

We've got your back, with picks you can turn to if you're ready to expand your collection or establish it in the first place. We've rounded up some of the best Pokémon TC deals you can shop as of Jan. 20. If you're ready to crack open some packs, consider these deals your gateway to doing so and saving some cash at the same time. Keep checking back, as we'll continue to add new deals that you won't want to miss out on.

Best Pokémon TCG Elite Trainer Box deal Opens in a new window Credit: Amazon Pokémon TCG Phantasmal Flames Elite Trainer Box $79.94 at Amazon
$106.89 Save $26.95   Get Deal Why we like it

Elite Trainer Boxes are the products you always want to look out for when it comes to every expansion, and when you can save some money on one, it's always a good idea to do so. This box comes with nine booster packs, a full-art promo card with Charcadet, dice, 65 special sleeves, and all the things you need to actually play the game. You get a total of 40 TCG Energy cards and a player's guide as well, with a plastic coin, a collector's box, and 6 dividers to keep it all organized. You'll also get a code for Pokémon TCG Live. As Phantasmal Flames cards continue to become even more expensive, this kind of access to them at an affordable price continues to shrink, so this is a good buy to snag while you still can.

Best Pokémon TCG booster bundle deal Opens in a new window Credit: Amazon Pokémon TCG: Mega Evolution Phantasmal Flames Booster Bundle $52.70 at Amazon
$64.99 Save $12.29   Get Deal Why we like it

Though the Phantasmal Flames expansion is no longer the newest addition to the Pokémon TCG, it's still extremely valuable. This booster bundle nets you six Phantasmal Flames booster packs, which are still priced around $10 apiece on Amazon. That means you'l be saving nearly the price of one pack in its entirety by buying this box than you would with individual boosters. Given the fact that this expansion contains the Mega Hyper Rare Charizard ex, which is selling up to $400 as of TCGPlayer, you could come away with some serious high-value cards, so snap up a couple of these boxes and you'll get several chances to pull it.

Best Pokémon TCG: Collector Chest deal Opens in a new window Credit: Amazon Pokémon TCG: Collector Chest Fall 2025 $37.99 at Amazon
$57.99 Save $20   Get Deal Why we like it

This fun collector chest comes in a special illustrated tin, which already makes it eye-catching and lucrative if you enjoy Pokémon as a fan or just as someone who likes special sets. And of course, saving $20 on the set is a good reason to be excited about it as well. This chest comes with six booster packs: three Mega Evolution, two Journey Together, and one Destined Rivals. Additionally, you get three foil promo cards of Ralts, Kirlia, and Riolu as well as a Mega Lucario coin, four sticker sheets, and a mini portfolio for your cards. All this lets you savor a great deal as well as some fun goodies without having to pay $50 to $60 for a booster bundle or booster box. Plus, you can store it all in the same place you got all the goodies.

More Pokémon TCG deals

• Pokémon TCG: Paldean Fates Great Tusk ex & Iron Threads ex Premium Collection — $174.93 $219.99 (save $45.06)

• Pokémon TCG: Mega Charizard X ex Ultra-Premium Collection — $149.99 $249.99 (save $100)

• Pokémon TCG: Holiday Calendar (2025) — $69.99 $74.89 (save $4.90)

• Pokémon TCG Scarlet & Violet 10.5 Black Bolt Binder Box — $62.94 $67.89 (save $4.95)

• Pokémon TCG: Mega Lucario ex Figure Collection — $58.94 $67.99 (save $6.91)

• Pokémon TCG: Charizard ex Special Collection — $57.99 $64.90 (save $6.91)

• Pokémon TCG: Trainer’s Toolkit — $34.95 $56.90 (save $22.04)

• Pokémon TCG: Team Rocket’s Mewtwo ex League Battle Deck — $19.79 $52.99 (save $33.20)

RAM prices have skyrocketed over the past few months, and it doesn’t look like they’re coming down anytime soon. With upgrades threatening to hurt your wallet more than ever, maybe it’s time to take a different approach: making smarter use of the RAM you already have.

At a glance

• Today’s multimodal AI systems can give answers that sound right but may not be grounded in what they actually observe over time, leading to unpredictable errors and safety risks in real-world settings.
• Argos is a verification framework for multimodal reinforcement learning that trains models by rewarding not just correct answers, but correct answers grounded in visual and temporal evidence, using automated verification rather than human labeling. It selects the appropriate specialized tools for each answer based on what needs to be verified. 
• Models trained with Argos show stronger spatial reasoning, far fewer visual hallucinations, more stable learning dynamics, and better performance on robotics and real-world tasks while requiring fewer training samples.

Over the past few years, AI systems have become much better at discerning images, generating language, and performing tasks within physical and virtual environments. Yet they still fail in ways that are hard to predict and even harder to fix. A robot might try to grasp a tool when the object is visibly blocked, or a visual assistant integrated into smart glasses might describe objects that aren’t actually present.

These errors often arise because today’s multimodal agents are trained to generate outputs that are plausible rather than grounded in the actual information they receive from their environment. As a result, a model’s output can seem correct while relying on incorrect information. As AI systems are increasingly used to navigate 3D spaces and make decisions in real-world settings, this gap can be a safety and reliability concern.

To tackle this challenge, we posed the question: How can we train AI agents to generate correct answers and take appropriate actions for the right reasons so that their behavior is reliable even as the environment or tasks change?

Argos represents a novel answer to this challenge. It’s an agentic verification framework designed to improve the reliability of reinforcement learning in multimodal models. Reinforcement learning is a training method where AI models learn by receiving rewards for desired behaviors and penalties for undesired ones, gradually improving their performance through trial and error.

Rather than rewarding only correct behaviors, Argos evaluates how those behaviors were produced. It draws on a pool of larger, more capable teacher models and rule-based checks to verify two things: first, that the objects and events a model references actually exist in its input, and second, that the model’s reasoning aligns with what it observes. Argos rewards the model when both conditions are met. In practice, these rewards help curate high-quality training data and guide the model’s further training.

How Argos works

Argos functions as a verification layer on top of an existing multimodal model. Given an image or video, a task or query, and information about the model’s reasoning and output, Argos identifies where the model indicates objects are located in the image, when it indicates events occur in a video, and what action or answer it produces.

Argos then applies specialized tools tailored to the specific content to evaluate and score three aspects of the model’s output. It checks whether the answer is correct, whether referenced objects and events appear at the indicated locations and times, and whether the reasoning is consistent with the visual evidence and the answer (Figure 1).

These scores are combined using a gated aggregation function, a method that dynamically adjusts the importance of different scores. It emphasizes reasoning checks only when the final output is correct. This design prevents unreliable feedback from dominating training and produces a stable reward signal for reinforcement learning.

Figure 1. Argos selects different specialized tools to verify and score the accuracy of referenced points and events in the agent’s reasoning. Using Argos to curate data for supervised fine-tuning

Argos also helps curate high-quality training data to provide the model with a strong foundation in grounded reasoning. Before the reinforcement learning stage begins, Argos uses a multi-stage process to generate data that is explicitly tied to visual locations and time intervals.

In the first stage, Argos identifies the objects, actions, and events that are relevant to a task and links them to specific locations in images or specific moments in videos. These references are overlaid on images and selected video frames. Next, a reasoning model generates step-by-step explanations that refer to these visual locations and time spans.

Finally, Argos evaluates each generated example for accuracy and visual grounding, filtering out low-quality training data and retaining only data that is both correct and well-grounded in visual input. The resulting dataset is then used in an initial training phase, where the model learns to generate reasoning steps before producing its final output. This process is illustrated in Figure 2.

Figure 2. Argos generates step-by-step reasoning grounded in image locations and video timestamps then filters out low-quality training data. Evaluation

Building on this foundation in grounded reasoning, we further trained the model using reinforcement learning guided by Argos and evaluated its performance across a range of benchmarks. On spatial reasoning tasks, the Argos-trained model outperformed both the base model Qwen2.5-VL-7B and the stronger Video-R1 baseline across challenging 3D scenarios and multi-view tasks. Models trained with Argos also showed a substantial reduction of hallucinations compared with both standard chain-of-thought prompting and reinforcement learning baselines.

Finally, we evaluated the model in robotics and other real-world task settings, focusing on high-level planning and fine-grained control. Models trained with Argos performed better on complex, multi-step tasks. Notably, these improvements were achieved using fewer training samples than existing approaches, highlighting the importance of reward design in producing more capable and data-efficient agents. Figure 3 illustrates some of these findings.

Figure 3. Performance of Argos compared with baseline models on the task of visual hallucination detection (left) and embodied task planning and completion (right).  How Argos shapes reinforcement learning

To understand how Argos affects learning, we took the same vision-language model that had been trained on our curated dataset and fine-tuned it using reinforcement learning in two different ways. In one approach, Argos was an agentic verifier, checking the correctness of outputs and the quality of reasoning. In the other, the model received feedback only on whether its answers were correct.

We evaluated both versions on 1,500 samples from a new dataset and tracked their performance throughout the learning process (Figure 4). Although they started at similar levels, the model without Argos quickly got worse. Its accuracy steadily declined, and it increasingly gave answers that ignored what was in the videos. It learned to game the system by producing answers that seemed correct without grounding them in visual evidence.

The model trained with Argos showed the opposite pattern. Accuracy improved steadily, and the model became better at linking its reasoning to what appeared in the videos. This difference highlights the value of verification: when training rewards both correct outputs and sound reasoning based on visual and temporal evidence, models learn to be more reliable rather than simply finding shortcuts to high scores.

Figure 4. Comparison of response accuracy changes with and without Argos across two model versions (left) and differences in visual grounding accuracy over training for both versions (right). Potential impact and looking forward

This research points toward a different way of building AI agents for real-world applications. Rather than fixing errors after they occur, it focuses on training agents to systematically anchor their reasoning in what they actually receive as input throughout the training process.

The potential applications span many domains. A visual assistant for a self-driving car that verifies what’s actually in an image is less likely to report phantom obstacles. A system that automates digital tasks and checks each action against what’s displayed on the screen is less likely to click the wrong button.

As AI systems move beyond research labs into homes, factories, and offices, reliable reasoning becomes essential for safety and trust. Argos represents an early example of verification systems that evolve alongside the AI models they supervise. Future verifiers could be tailored for specific fields like medical imaging, industrial simulations, and business analytics. As more advanced models and richer data sources become available, researchers can use them to improve these verification systems, providing even better guidance during training and further reducing hallucinations.

We hope that this research helps move the field toward AI systems that are both capable and interpretable: agents that can explain their decisions, point to the evidence behind them, and be trained to adhere to real-world requirements and values.

Opens in a new tab

The post Multimodal reinforcement learning with agentic verifier for AI agents appeared first on Microsoft Research.