TikTok’s major commerce investment in South East Asia has hit some speed bumps.
House Speaker Mike Johnson says he’ll look to move forward with three crypto bills on Wednesday after some Republican lawmakers pulled support over wanting a CBDC ban.
xAI’s first AI companions on the Grok app are a lustful anime girl and a homicidal panda.
Read more The fund brings Mantis’ assets under management to $225 million.
Read more Amazon is looking to its main competitor to get its next batch of Kuiper satellites to orbit.
Read more Cameron Wagenius had already pleaded guilty to other charges earlier this year.
Read more A senior engineer describes the thrill, chaos, and exhaustion of working at OpenAI this past year while launching its coding agent Codex.
Read more In June 2023, technology leaders and IT services executives had a lightning bolt headed their way when McKinsey published the “The economic potential of generative AI: The next productivity frontier” report. It echoed a moment from the 2010s when Amazon Web Services launched an advertising campaign aimed at Main Street’s C-suite: Why would any fiscally responsible exec allow their IT teams to spend capex for servers and software when AWS only cost 10 cents per virtual machine?
Vendors understand that these kinds of reports and aggressive advertising around competitive risks projected onto an industry sector would drive many calls from boards to their C-suite, rolling from C-suite to their staff all asking, “What are we doing with AI?” When asked to “do something with AI,” technical leadership and their organizations promptly responded — sometimes begrudgingly and sometimes excitedly — for work-sanctioned opportunities to get their hands on a new technology. At that point, there was no time to sort between actual business returns from applying AI and “AI novelty” use cases that were more Rube Goldberg machines than tangible breakthroughs.
Today’s opportunity: Significant automation gains
When leaders respond to immediate panic, new business risks and mitigations often emerge. Two recent examples highlight the consequences of rushing to implement and publish positive results from AI adoption. The Wall Street Journal reported in April 2025 on companies struggling to realize returns on AI. Just weeks later, it covered MIT’s retraction of a technical paper about AI where the results that led to its publication could not be substantiated.
While these reports demonstrate the pitfalls of over-reliance on AI without common-sense guardrails, not all is off track in the land of enterprise AI adoption. Incredible results being found from judicious use of AI and related technologies in automating processes across industries. Now that we are through the “fear of missing out” stage and can get down to business, where are the best places to look for value when applying AI to automation of your business?
While chatbots are almost as pervasive as new app downloads for mobile phones, the applications of AI realizing automation and productivity gains line up with the unique purpose and architecture of the underlying AI system they are built on. The dominant patterns where AI gains are realized currently boil down to two things: language (translation and patterns) and data (new format creation and data search).
Example one: Natural language processing
Manufacturing automation challenge: Failure Mode and Effects Analysis (FMEA) is both critical and often labor intensive. It is not always performed prior to a failure in manufacturing equipment, so very often FMEA occurs in a stressful manufacturing lines-down scenario. In Intel’s case, a global footprint of manufacturing facilities separated by large distances along with time zones and preferred language differences makes this even more difficult to find the root cause of a problem. Weeks of engineering effort are spent per FMEA analysis repeated across large fleets of tools spread between these facilities.
Solution: Leverage already deployed CPU compute servers for natural language processing (NLP) across the manufacturing tool logs, where observations about the tools’ operations are maintained by the local manufacturing technicians. The analysis also applied sentiment analysis to classify words as positive, negative, or neutral. The new system performed FMEA on six months of data in under one minute, saving weeks of engineering time and allowing the manufacturing line to proactively service equipment on a pre-emptive schedule rather than incurring unexpected downtime.
Financial institution challenge: Programming languages commonly used by software engineers have evolved. Mature bellwether institutions were often formed through a series of mergers and acquisitions over the years, and they continue to rely on critical systems that are based on 30-year-old programming languages that current-day software engineers are not familiar with.
Solution: Use NLP to translate between the old and new programming languages, giving software engineers a needed boost to improve the serviceability of critical operational systems. Use the power of AI rather than doing a risky rewrite or massive upgrade.
Example two: Company product specifications and generative AI models
Sales automation challenge: The time it takes to reformat a company’s product data into a specific customer RFP format has been an ongoing challenge across industries. Teams of sales and technical leads spend weeks of work across different accounts reformatting the same root data between the preferred PowerPoint or Word document formats. The customer response times were measured in weeks, especially if the RFPs required legal reviews.
Solution: By using generative AI combined with a data extraction and prompting technique called retrieval augmented generation (RAG), companies can rapidly reformat product information between different customer required RFP response formats. The time spent moving data between different documents and different document types only to find an unforced error in the move is reduced to hours instead of weeks.
HR policy automation challenge: Navigating internal processes can be time consuming and confusing for both HR and employees. The consequences of misinterpretation, access outages, and personal information or private data being exposed are massively important to the company and the individual.
Solution: Combine generative AI, RAG, and an interactive chatbot that uses employee-assigned assets to determine identity and access rights, provides employees interactive query-based chat formats to answer their questions in real time.
Finding your best use cases for AI
In a world where 80% to 90% of all AI proof of concepts fail to scale, now is the time to develop a framework that is based on caution. Consider starting with a data strategy and governance assessment. Then find opportunities to compare successful AI-based automation efforts at peer companies through peer discussions. Clear, rules-based policies and processes offer the best opportunities to begin a successful AI automation journey in your enterprise. Where you encounter disparate data sources (e.g., unstructured, video, structured databases) or unclear processes, maintain tighter human-in-the-loop decision controls to avoid unexpected data or token exposure and cost overruns.
As the AI hype cycle cools and business pressure mounts, now is the time to become practical. Apply AI to well-defined use cases and begin unlocking the automation benefits that will matter not just in 2025, but for years to come.
This content was produced by Intel. It was not written by MIT Technology Review’s editorial staff.
Adaptive production is more than a technological upgrade: it is a paradigm shift. This new frontier enables the integration of cutting-edge technologies to create an increasingly autonomous environment, where interconnected manufacturing plants go beyond the limits of traditional automation. Artificial intelligence, digital twins, and robotics are among the powerful tools manufacturers are using to create dynamic, intelligent systems that not only perform tasks, but also learn, make decisions, and evolve in real-time.
Taking this kind of adaptive approach can transform a manufacturer’s productivity, efficiency, and innovation. But beyond the factory, it also has the potential to deliver society-wide benefits, by bolstering economic growth locally, creating more attractive and accessible employment opportunities, and supporting a sustainability agenda.
As efforts to revive and modernize local manufacturing accelerate in regions around the world, including North America and Europe, adaptive production could help manufacturers overcome some of their biggest obstacles—firstly, attracting and retaining talent. Nearly 60% of manufacturers cited this as their top challenge in a 2024 US-based survey. Highly automated, technology-led adaptive production methods hold new promise for attracting talent to roles that are safer, less repetitive, and better paid. “The ideal scenario is one where AI enhances human capabilities, leads to new task creation, and empowers the people who are most at risk from automation’s impact on certain jobs, particularly those without college degrees,” says Simon Johnson, co-director of MIT’s Shaping the Future of Work Initiative.
Secondly, the digitalization of manufacturing—embedded in the very foundation of adaptive production technologies—allows companies to better address complex sustainability challenges through process and resource optimization and a better understanding of data. “By integrating these advanced technologies, we gain a more comprehensive picture across the entire production process and product lifecycle,” explains Jelena Mitic, head of technology for the Future of Automation at Siemens. “This will provide a much faster and more efficient way to optimize operations and ensure that all the necessary safety and sustainability requirements are met during quality control.”
This content was produced by Insights, the custom content arm of MIT Technology Review. It was not written by MIT Technology Review’s editorial staff.
This content was researched, designed, and written entirely by human writers, editors, analysts, and illustrators. This includes the writing of surveys and collection of data for surveys. AI tools that may have been used were limited to secondary production processes that passed thorough human review.
As soon as Google launched its latest video-generating AI model at the end of May, creatives rushed to put it through its paces. Released just months after its predecessor, Veo 3 allows users to generate sounds and dialogue for the first time, sparking a flurry of hyperrealistic eight-second clips stitched together into ads, ASMR videos, imagined film trailers, and humorous street interviews. Academy Award–nominated director Darren Aronofsky used the tool to create a short film called Ancestra. During a press briefing, Demis Hassabis, Google DeepMind’s CEO, likened the leap forward to “emerging from the silent era of video generation.”
But others quickly found that in some ways the tool wasn’t behaving as expected. When it generates clips that include dialogue, Veo 3 often adds nonsensical, garbled subtitles, even when the prompts it’s been given explicitly ask for no captions or subtitles to be added.
Getting rid of them isn’t straightforward—or cheap. Users have been forced to resort to regenerating clips (which costs them more money), using external subtitle-removing tools, or cropping their videos to get rid of the subtitles altogether.
Josh Woodward, vice president of Google Labs and Gemini, posted on X on June 9 that Google had developed fixes to reduce the gibberish text. But over a month later, users are still logging issues with it in Google Labs’ Discord channel, demonstrating how difficult it can be to correct issues in major AI models.
Like its predecessors, Veo 3 is available to paying members of Google’s subscription tiers, which start at $249.99 a month. To generate an eight-second clip, users enter a text prompt describing the scene they’d like to create into Google’s AI filmmaking tool Flow, Gemini, or other Google platforms. Each Veo 3 generation costs a minimum of 20 AI credits, and the account can be topped up at a cost of $25 per 2,500 credits.
Mona Weiss, an advertising creative director, says that regenerating her scenes in a bid to get rid of the random captions is becoming expensive. “If you’re creating a scene with dialogue, up to 40% of its output has gibberish subtitles that make it unusable,” she says. “You’re burning through money trying to get a scene you like, but then you can’t even use it.”
When Weiss reported the problem to Google Labs through its Discord channel in the hopes of getting a refund for her wasted credits, its team pointed her to the company’s official support team. They offered her a refund for the cost of Veo 3, but not for the credits. Weiss declined, as accepting would have meant losing access to the model altogether. The Google Labs’ Discord support team has been telling users that subtitles can be triggered by speech, saying that they’re aware of the problem and are working to fix it.
So why does Veo 3 insist on adding these subtitles, and why does it appear to be so difficult to solve the problem? It probably comes down to what the model has been trained on.
Although Google hasn’t made this information public, that training data is likely to include YouTube videos, clips from vlogs and gaming channels, and TikTok edits, many of which come with subtitles. These embedded subtitles are part of the video frames rather than separate text tracks layered on top, meaning it’s difficult to remove them before they’re used for training, says Shuo Niu, an assistant professor at Clark University in Massachusetts who studies video sharing platforms and AI.
“The text-to-video model is trained using reinforcement learning to produce content that mimics human-created videos, and if such videos include subtitles, the model may ‘learn’ that incorporating subtitles enhances similarity with human-generated content,” he says.
“We’re continuously working to improve video creation, especially with text, speech that sounds natural, and audio that syncs perfectly,” a Google spokesperson says. “We encourage users to try their prompt again if they notice an inconsistency and give us feedback using the thumbs up/down option.”
As for why the model ignores instructions such as “No subtitles,” negative prompts (telling a generative AI model not to do something) are usually less effective than positive ones, says Tuhin Chakrabarty, an assistant professor at Stony Brook University who studies AI systems.
To fix the problem, Google would have to check every frame of each video Veo 3 has been trained on, and either get rid of or relabel those with captions before retraining the model—an endeavor that would take weeks, he says.
Katerina Cizek, a documentary maker and artistic director at the MIT Open Documentary Lab, believes the problem exemplifies Google’s willingness to launch products before they’re fully ready.
“Google needed a win,” she says. “They needed to be the first to pump out a tool that generates lip-synched audio. And so that was more important than fixing their subtitle issue.”