The Godfather of AI lays out a key difference between OpenAI and Google when it comes to safety
When it comes to winning the AI race, the "Godfather of AI" thinks there's an advantage in having nothing to lose.
On an episode of the "Diary of a CEO" podcast that aired June 16, Geoffrey Hinton laid out what he sees as a key difference between how OpenAI and Google, his former employer, dealt with AI safety.
"When they had these big chatbots, they didn't release them, possibly because they were worried about their reputation," Hinton said of Google. "They had a very good reputation, and they didn't want to damage it."
Google released Bard, its AI chatbot, in March of 2023, before later incorporating it into its larger suite of large language models called Gemini. The company was playing catch-up, though, since OpenAI released ChatGPT at the end of 2022.
Hinton, who earned his nickname for his pioneering work on neural networks, laid out a key reason that OpenAI could move faster on the podcast episode: "OpenAI didn't have a reputation, and so they could afford to take the gamble."
Talking at an all-hands meeting shortly after ChatGPT came out, Google's then-head of AI said the company didn't plan to immediately release a chatbot because of " reputational risk," adding that it needed to make choices "more conservatively than a small startup," CNBC reported at the time.
The company's AI boss, Google DeepMind CEO Demis Hassabis, said in February of this year that AI poses potential long-term risks, and that agentic systems could get "out of control." He advocated having a governing body that regulates AI projects.
Gemini has made some high-profile mistakes since its launch, and showed bias in its written responses and image-generating feature. Google CEO Sundar Pichai addressed the controversy in a memo to staff last year, saying the company " got it wrong" and pledging to make changes.
The " Godfather" saw Google's early chatbot decision-making from the inside — he spent more than a decade at the company before quitting to talk more freely about what he describes as the dangers of AI. On Monday's podcast episode, though, Hinton said he didn't face internal pressure to stay silent.
"Google encouraged me to stay and work on AI safety, and said I could do whatever I liked on AI safety," he said. "You kind of censor yourself. If you work for a big company, you don't feel right saying things that will damage the big company."
Overall, Hinton said he thinks Google "actually behaved very responsibly."
Hinton couldn't be as sure about OpenAI, though he has never worked at the company. When asked whether the company's CEO, Sam Altman, has a "good moral compass" earlier in the episode, he said, "We'll see." He added that he doesn't know Altman personally, so he didn't want to comment further.
OpenAI has faced criticism in recent months for approaching safety differently than in the past. In a recent blog post, the company said it would only change its safety requirements after making sure it wouldn't "meaningfully increase the overall risk of severe harm." Its focus areas for safety now include cybersecurity, chemical threats, and AI's power to improve independently.
Altman defended OpenAI's approach to safety in an interview at TED2025 in April, saying that the company's preparedness framework outlines "where we think the most important danger moments are." Altman also acknowledged in the interview that OpenAI has loosened some restrictions on its model's behavior based on user feedback about censorship.
The earlier competition between OpenAI and Google to release initial chatbots was fierce, and the AI talent race is only heating up. Documents reviewed by Business Insider reveal that Google relied on ChatGPT in 2023 — during its attempts to catch up to ChatGPT.
Try Our AI Features
Explore what Daily8 AI can do for you:
Comments
No comments yet...
Related Articles
Business Insider
15 minutes ago
- Business Insider
A massive trove of 16 billion stolen passwords was discovered — here's what to do
Researchers say they've uncovered one of the largest data leaks in history that involves many popular platforms. The leak includes nearly 16 billion login credentials that could give cybercriminals access to social media and business platforms such as Apple, Gmail, Telegram, Facebook, GitHub, and more, researchers at Cybernews said this week. Bad actors now have "unprecedented access to personal credentials that can be used for account takeover, identity theft, and highly targeted phishing," the researchers said. The number of exposed people or accounts is unknown. The researchers said the data likely comes from malicious software known as infostealers. "What's especially concerning is the structure and recency of these datasets — these aren't just old breaches being recycled. This is fresh, weaponizable intelligence at scale," the researchers said. Cybernews said researchers uncovered the leak when the datasets were exposed for a short period of time. It follows the May discovery of a database containing more than 184 million credentials, including Apple, Facebook, and Google logins, Wired earlier reported. If you're nervous that your logins are at risk, there are steps you can take to make your account safer. How to protect yourself You can't unring the bell of an information leak. However, you can take steps to identify if your credentials have been involved in any data breaches and protect yourself in the future. You can check sites like Have I Been Pwned to see if your email has appeared in a data breach. Turning on two-step authentication for your accounts can also help protect them from unauthorized access. Platforms also offer resources to help users secure their accounts. Google encourages users to use protections that don't require a password, like a passkey. It's one of the tech giants, along with Apple, Amazon, and Microsoft, that have been working to move users away from passwords to help secure their accounts. For those who prefer to stick with passwords, Google's password manager can store login credentials and notify users if they appear in a breach, a spokesperson told Business Insider. There's also Google's dark web report, a free tool that tracks whether personal information is floating around in online databases. GitHub, an online coding platform, offers developers a guide on how to implement safety measures in their organizations. The site recommends creating a security policy, having strict password guidelines, and requiring two-factor authorization. The data leak included logs — "often with tokens, cookies, and metadata," which makes it "particularly dangerous for organizations lacking multi-factor authentication or credential hygiene practices," the Cybernews team said. Meta offers a Privacy Checkup tool for users to review their privacy and security account settings. There, you can turn on two-factor authentication and ensure Meta alerts you of unusual logins. Meanwhile, Telegram said its primary login method sends a one-time password to users over SMS. "As a result, this is far less relevant for Telegram users compared to other platforms where the password is always the same," a Telegram spokesperson told BI about the data leak. Apple, GitHub, and Meta did immediately respond to a request for comment on the data leak. Google said it was directing users to some of the security resources above.
Forbes
30 minutes ago
- Forbes
Cisco Live 2025 Touts Cisco's Platform Advantages For Enterprise AI
Cisco president and chief product officer Jeetu Patel presents at Cisco Live 2025. At last week's Cisco Live in San Diego, CEO Chuck Robbins said that it would be the most important Cisco Live ever and announce more innovations than ever. Having attended a ton of these events and followed the company closely for many years, I can tell you that the show mostly fulfilled that promise. More than that, it reinforced Cisco's areas of strategic focus in infrastructure, the modern workplace and digital resiliency. It's clear that Cisco is working hard to leverage its platform advantages across networking, security, observability, compute and even silicon to support agentic AI workloads. This should help customers simplify operations while maintaining the highest levels of network security and AI safety. As I pointed out in my analysis of the Cisco Partner Summit held late in 2024, Cisco was very deliberate — perhaps a touch slow — in establishing its overarching AI strategy. By now, though, I'm impressed with how quickly the company has been moving to bring this strategy to fruition. Let's dig into the details of what Cisco is doing, and what it could still do better. (Note: Cisco is an advisory client of my firm, Moor Insights & Strategy.) Cisco's Strategic Imperatives, Per Robbins And Patel Early on, Robbins made the fundamental point that networking is critical for AI to function, and will be a big factor in enabling AI growth going forward. Beyond that, agentic AI will be adequately secured only by applying security to the network. In this context, Robbins scored a direct hit on Cisco's competitors when he pointed out that 'None of our network friends have security and none of our security friends have networking.' He believes (and I agree) that this puts Cisco in a unique position to help integrate security into the network, which I think is going to be especially important in enterprise IT. AI is growing like wildfire against a backdrop of global turmoil. Robbins said that geopolitical dynamics are a big concern for Cisco, noting that AI competition isn't just between companies, but between nations as well. Whether companies or countries, everyone has FOMO, and everyone feels like they need to move fast. (He said that 85% of enterprises believe they must 'do AI' in the next 18 months.) This reminds me that most of my conversations at this year's World Economic Forum in Davos were spent discussing countries' needs for a disconnected, sovereign AI cloud. The need is there. A new type of infrastructure will be required to realize the potential of generative and especially agentic AI. According to Robbins, we need a similar kind of advancement in infrastructure that happened when the internet became ubiquitous in the 1990s. It's worth pausing here to reflect on how important that period of technology was for turning Cisco into a networking juggernaut. Here in the 2020s, this is going to play out with a hybrid strategy that includes both cloud hyperscalers and private enterprise datacenters. For meeting this need, Robbins told the crowd not to underestimate the impact of Cisco's combined strengths in networking, security and silicon. (In my view, Cisco needs to talk more often and in more detail about the silicon part. I'll come back to that in my recommendations at the end of this article.) After the CEO's keynote was done, Cisco's chief product officer — and newly appointed president — Jeetu Patel took the stage and echoed Robbins with his emphasis on: He later gave much more detail on each of those facets, as I'll cover below. But first he talked about how the ability of AI agents to autonomously execute tasks will compound productivity, especially when combined with advances in robotics, AI and other areas. As he put it, '8 billion [people] will feel like 80 billion.' However, this productivity explosion will be constrained by limits in power, networking and compute. He also foresees a growing divide between companies that are dexterous with AI and those that will struggle. 'We want to help you be in that first category,' Patel said. He's making a very timely prediction. I presented to a group of European CIOs earlier this week in Munich, and one of the slides showed logos of companies that 'died' from not embracing the internet and e-commerce. The same will happen to companies that don't quickly embrace AI. You don't have to be first, but you can't be last. How can Cisco help? Patel brought it back to the compounding effect of Cisco's platform approach, where many different types of complementary technology work 'in harmony.' He referenced the company's silicon (so customers aren't stuck with a single provider) and especially programmable silicon (to adapt to new use cases). He also reiterated a point he had made in the 2024 Cisco Partner Summit — that AI is foundational to Cisco's products, so customers can expect it to be built right in. While I think that's an accurate thing to say, I would also suggest that by now it's not such a point of differentiation. The Need For AI-Ready Infrastructure Patel went into more detail about the massive, even exponential, buildout of datacenters underway right now. He said that Cisco is foundational in building out these new datacenters. For datacenters to support large-scale agentic AI, they need a new architecture that can support the constant high levels of AI model activity that agentic creates. This is unlike generative AI chatbots, where the activity spikes up and down. Patel believes that the company can take advantage of the opportunity based on the experience it has gained from many years of serving hyperscalers/CSPs, neoclouds and enterprise customers. In support of my praise for Cisco's impressive speed lately, Patel touted the 19 major datacenter innovations the company has launched just within the past six months. At Cisco Live, it announced the unified Nexus Dashboard, which creates 'one brain for all of our data center fabrics,' according to Patel. There was also plenty of talk about the company's partnership with AI bellwether Nvidia. Among other aspects of the pairing, Cisco switches are completely integrated into Nvidia architecture, and Nvidia NeMo models can be secured with Cisco AI Defense. As I have said before, I am a recovering product management and product marketing executive, and I always challenge tech companies to describe their product realization process. While Cisco gets criticized for its 'legacy' roots, Patel has very much changed the product culture there. Fewer layers and faster time-to-decision. Most of the new software underlying AI was developed by small teams with six to eight members. This is a new practice — and very much a new Cisco. I will be digging more into the metrics and outcomes as they're available, but I like what I hear so far. Given that this is Cisco we're talking about, that was just the tip of the iceberg for cybersecurity. Patel described security as a prerequisite for enterprise AI because 'If people don't trust the system, they're not going to use it.' There was also an announcement about the Hybrid Mesh Firewall, which enables distributed policy enforcement, adds security to all sorts of devices and can work with existing firewalls (even from third parties). There were other announcements of specific firewalls, and Patel asserted that Cisco is the price-performance leader for firewalls at every level of scale. The company also launched a new secure network architecture called Cisco Live Protect, which is meant to shield your network from an exploit within minutes to give your IT security team time to fix the underlying issue. The contrast between 'within minutes' and the industrywide 45-day average to patch a vulnerability is striking, to say the least. You can read more in this analysis from my colleague Will Townsend, who's an expert on networking and cybersecurity. Our colleague Matt Kimball, who has a long background in datacenters, will also be publishing his analysis soon. Networking And Equipping The Workplace Of Tomorrow This part of the presentation bridged various aspects of networking for enterprises, where Patel said the priorities were operational simplicity, scalability and — once again — security infused into the network. He got some cheers when he announced that Cisco's Catalyst switches are now unified with its Meraki network platform; there's now a single dashboard for managing these along with all of Cisco's next-gen devices. From my perspective, this is a nice example of Cisco's growing emphasis on easing the customer/user experience. In that vein, there was also an impressive demo of the new AgenticOps platform, which includes a multiplayer management console called AI Canvas. Will Townsend wrote much more about this in his article, praising its 'dynamic and real-time view into the inner workings of a customer's infrastructure expanse' to manage the network assurance, observability and remediation supplied by other Cisco tools. The live demo showed a user fetching data on a network outage and making UI widgets in real time to manage it. The engineer using it walked through troubleshooting, then inviting other users to help — with an autogenerated AI summary of what had been done so far. The AI model recognized missing data and looked for it, and then it was easy to apply a patch straight from the dashboard. Even a non-engineer like me could see immediately how helpful this console would be. Patel is not afraid to use hyperbole when it's warranted, so he summarized the impact of AgenticOps by saying, 'The way in which you run your network will never be the same again.' And he promised that much more innovation like this is coming through the pipeline. There was a lot more, including 'one of the largest refreshes of networking devices in Cisco history.' This includes smart switches, secure routers, WiFi 7 gear, campus gateways, industrial IoT . . . if you can network it, Cisco wants to do it smarter. For example, the new smart switches have isolated compute so you can run things like security right on the switch, plus all of the devices act as sensors that provide information about their environment back to the system. Harnessing Data And AI To Fortify Digital Resilience When the conversation turned to digital resilience, Patel and other presenters continued the theme of bringing more data into the picture to keep infrastructure running well. Enterprises routinely expend many hours determining the causes of outages; in Patel's view, the friction of this process is created by not having the right data available. 'One of the reasons we acquired Splunk for the low price of $28 billion,' he said, 'was to take all this data across multiple domains and correlate it.' He added that the core method of digital resilience is to distill data, correlate it, then unleash AI on the problem. There were plenty of specifics in terms of new launches, new Splunk integrations and so on, not to mention using smaller, more efficient bespoke AI models for specific security needs. (Those Cisco folks really are building AI into everything.) But for me there were two big takeaways from this part of the show. First is the idea of reimagining security operations by performing security at machine scale, and consolidating and simplifying security solutions to make that easier. Second is the extension of observability to AI. Some aspects of AI may still be 'black boxes' in terms of what the algorithms are doing, but Cisco wants to give its customers the ability to see everything their AIs are doing in terms of compute usage, network traffic, power draw and so on. If the company is able to pull off everything talked about onstage, I think that can only help with operationalizing AI to yield real business results for enterprises. Messaging All This AI Innovation Cisco Live reminds me a little bit of Google Cloud Next in terms of the sheer number of announcements packed into a couple of days. And it makes sense, given that both Cisco and Google Cloud (a) operate across multiple product areas with many different individual offerings in them, and (b) are investing enormous amounts of bandwidth (and capex) into AI — at speed. There are a couple of risks I see for Cisco. First, yes, the company's leaders told us repeatedly that AI is 'foundational' in their products. But it's one thing to claim this — as do most of the big infrastructure providers and enterprise software vendors. And the demos were great. But it's another thing to accurately engage with the market as a whole and with individual customers to help them understand the many potential payoffs for Cisco's AI innovations, area by area and product by product. I have great faith in Cisco's go-to-market prowess, but there's so much coming down the pike so fast that, to be most successful, the company needs to do the very best job of explaining its wares that it's ever done. Maybe it's my background in semiconductors talking, but I see this especially in how Cisco talks about its in-house silicon. In my view, Cisco doesn't blow this trumpet as loudly or as often as it should. The company announced its Silicon One initiative in 2019, and has been shipping its own chips for years by this point . . . yet there still aren't enough people who know about it. And look at the valuations of AI chip companies today. So I urge the company to talk more about its chips and how they add to the differentiation of Cisco's portfolio. The good news is, the introduction of all this AI functionality from Cisco — and the sea change underway in AI datacenter infrastructure — offers a perfect opportunity to do this. Cisco has some unique advantages in the market, starting with Chuck Robbins' correct and fundamental assertion that no one else in networking can match Cisco in security, and no one else in cybersecurity can match Cisco for networking. I'm also impressed by the waves of smart, relevant, user-friendly products I see at each Cisco event I attend. Now I want to see just how well Cisco can market, message and sell all this goodness into the enterprise.
TechCrunch
30 minutes ago
- TechCrunch
Startups Weekly: Fast and furious
Welcome to Startups Weekly — your weekly recap of everything you can't miss from the world of startups. Want it in your inbox every Friday? Sign up here. Some startups accrued value at lightning speed this week, and we got confirmation that defense tech is red hot. Most interesting startup stories from the week Image Credits:Kelly Sullivan / Getty Images Many startup stories this week occurred in Y Combinator's orbit in some way. Also, Israel once again lived up to its 'Start-Up Nation' reputation. That was fast: No-code website-building platform Wix acquired 6-month-old, bootstrapped vibe-coding startup Base44 — both Israeli companies — for $80 million in cash. That was fast, too: In just three months, Ramp's valuation jumped to $16 billion following its Series E, up from $13 billion when the spend management startup did a secondary sale earlier this year. Friends and foes: New details emerged on Meta's $14.3 billion deal to acquire 49% of startup Scale AI, including a potential dividend payout. We also learned that OpenAI was dropping Scale AI as a data provider following the deal. Frenemies: The U.S. Department of Defense awarded a contract worth up to $200 million directly to OpenAI, which could further strain the startup's relationship with Microsoft. Techcrunch event Save $200+ on your TechCrunch All Stage pass Build smarter. Scale faster. Connect deeper. Join visionaries from Precursor Ventures, NEA, Index Ventures, Underscore VC, and beyond for a day packed with strategies, workshops, and meaningful connections. Save $200+ on your TechCrunch All Stage pass Build smarter. Scale faster. Connect deeper. Join visionaries from Precursor Ventures, NEA, Index Ventures, Underscore VC, and beyond for a day packed with strategies, workshops, and meaningful connections. Boston, MA | REGISTER NOW ICYMI: Out of all the teams presenting at YC's recent Spring 2025 Demo Day, here are 11 startups that investors have been talking about. Didn't happen: People are also still talking about the police shutting down the YC Demo Day after-party that controversial AI startup Cluely tried to throw; or as its CEO told TechCrunch, 'the most legendary party that never happened.' Most interesting VC and funding news this week Helsing Image Credits:Helsing Most funding news was driven by either defense tech, AI, or both, but there were also some surprises. Plus, one VC firm is aiming high for its next fund. Good intuition: Applied Intuition, a company making software for autonomous vehicles, secured a $600 million Series F and tender offer at a $15 billion valuation. New tune: Munich-based defense tech startup Helsing closed a €600 million investment led by Prima Materia, the VC firm of Spotify's founder Daniel Ek, which valued Helsing at €12 billion. New unicorn: Israeli observability startup Coralogix became a unicorn after raising a $115 million Series E, which it will use to double its headcount in India, where 100 of its 550 employees are currently based. Toldja: Mach Industries, a 2-year-old defense tech startup, confirmed having raised a $100 million round of funding led by Khosla Ventures and Bedrock at a $470 million valuation. Money flows: Aspora, a startup formerly known as Vance and focused on facilitating remittances from the Indian diaspora, closed a $50 million Series B co-led by Sequoia and Greylock at a $500 million valuation. Change of heart: Sword Health, an AI-powered digital health startup that began as a virtual physical therapy solution, locked in $40 million at a $4 billion valuation in a funding round led by returning investor General Catalyst. It also pushed back its IPO plans to at least 2028. Multiplier effect: Multiplier Holdings announced having raised $27.5 million across seed and Series A rounds after joining the growing trend of buying legacy service businesses — in its case, accounting firms — and scaling them with AI. Stock where you shop: Grifin, a startup whose app helps users buy stock from brands they shop at, such as Walmart, secured $11 million in a Series A round. Out of Sweden: Polar, a payment infrastructure platform for developers and AI-first businesses, raised a $10 million seed round led by Accel. Its CEO previously co-founded Tictail, which was acquired by Shopify in 2018. Bullish: Global VC Endeavor Catalyst is seeking to raise $300 million for its fifth fund. This would be its largest yet, as it looks to deepen its bet on fast-growing startups in emerging markets. Last but not least Image Credits:Alexa von Tobel An early investor in Chime, founder-turned-VC Alexa von Tobel is ready for Fintech 3.0. 'The next wave of innovation won't come from superficial tweaks but from fundamental deep product reinvention — tools that meet the needs of a changing economy and a more diverse, digitally native population,' she told TechCrunch in an interview.
