The Tool Desk
Outbyte PC Repair FREEClear out junk files and repair common Windows errorsFree Scan →Outbyte Driver Updater FREEScan for outdated or missing drivers - takes under a minuteDriver Scan →Some links on this page are affiliate links: if you buy through them we may earn a commission, at no extra cost to you.
Gary Marcus differs from other prominent AI critics mainly in where he places the emphasis: he argues that today’s generative AI is already dangerous because unreliable, weakly grounded systems can scale misinformation and other harms, even if they are nowhere near artificial general intelligence (AGI). Geoffrey Hinton is more focused on the future challenge of controlling systems that might surpass human intelligence; Yann LeCun is more optimistic that engineering can improve current limits and rejects near-term extinction claims. Yoshua Bengio weighs capability gains against broad risks, while Emily Bender emphasizes documented present harms and warns against letting speculative scenarios distract policymakers.
What is Gary Marcus’s central criticism of AI?
Marcus’s signature position is that generative AI can be “stupid” and dangerous at the same time. By “stupid,” he means unreliable and weakly grounded—not incapable of producing useful work. By dangerous, he means that people and organizations can deploy and trust systems that fabricate convincing material, while bad actors can use them to scale manipulation and other social harms.
As an Amazon Associate I earn from qualifying purchases.
In an IEEE Spectrum profile published in 2024, Marcus described generative AI as something that “cannot be trusted or counted on” and yet is dangerous. His concern is that limitations do not prevent harm: weak grounding can make a system easy to manipulate, and fluent output can make falsehoods seem credible. The profile discusses his book Taming Silicon Valley: How We Can Ensure That AI Works for Us (MIT Press, 2023), which addresses mass-produced disinformation, deepfake pornography, unlicensed use of creative work, and regulation of AI companies.
Marcus is not an opponent of all AI. He sees bounded uses, such as autocomplete, and argues for a better direction rather than abandoning the field. His skepticism is aimed at the dominant technical and commercial trajectory: he rejects the idea that scaling current large language models (LLMs) means AGI is imminent, and he argues that rapid integration without adequate trustworthiness and oversight is risky.
#1 Best Overall
- Read Before You Buy — No Video Output: These adapters support charging and USB 2.0 data transfer, but cannot transmit video signals. Except for standard USB webcams (which use USB data only), they are not compatible with HDMI/DisplayPort cables, video-capable USB-C hubs, or docking stations with video output.
- Convert USB-A Ports to USB-C: Designed to connect USB-C earphones, cables, flash drives, card readers, and other USB-C accessories to standard USB-A ports. Plug-and-play with no drivers or software required.
- Aluminum Alloy Housing: Built with a sturdy aluminum alloy shell that aids in heat dissipation and protects against daily wear and scratches. Designed to maintain a stable and secure connection.
- Compact & Travel-Friendly: The ultra-compact design allows the adapter to stay plugged into your device without blocking adjacent ports or adding bulk, reducing wear and tear on your original USB ports.
- 12-Month Warranty: Backed by a 12-month manufacturer warranty for peace of mind. Designed to meet strict quality control standards for reliable everyday performance.
This combination distinguishes him from a critic who objects to AI in principle. Marcus’s objection is practical and institutional: a system that is unreliable in ordinary use can still be embedded in schools, workplaces, media pipelines or public communication, where errors and fabricated material travel farther than they would in a private experiment. He also questions whether market incentives reward careful deployment when companies are competing to integrate systems quickly.
Why does Marcus say AI can be both unreliable and dangerous?
His argument separates a system’s ability to produce persuasive output from its ability to ground that output reliably. A model can generate plausible text without consistently knowing whether its claims, sources, or apparent evidence are sound. That gap matters when people rely on outputs or when a malicious user exploits the system’s weaknesses.
Quick wins for a faster PC:
Repair Windows errors before they cause bigger problemsFix Now →Scan for outdated or missing drivers - takes under a minuteDriver Scan →Marcus connects the problem to the information environment, not only to individual incorrect answers. In a May 8, 2023 Calcalist interview, he warned that the direction of AI could lead to a world where people trust nothing, making democratic functioning more difficult. He pointed to combinations of deepfakes and language models that can fabricate convincing studies, data, and references. His immediate concern is therefore not only whether a model makes a mistake, but whether cheap, convincing fabrications make it harder to distinguish authentic information from manipulated material.
That reasoning also explains why Marcus does not treat present limitations as reassurance. A tool need not be generally intelligent to produce harmful outputs at scale. The danger can arise from deployment, incentives, overtrust, and misuse before any system approaches human-level general intelligence.
In this view, “hallucination” is not a minor cosmetic defect. When a fluent answer includes invented references or unsupported data, a user may not notice the failure, especially when the output is combined with realistic images or video. The social cost can be cumulative: repeated fabrications can erode confidence in genuine records, journalism and expert testimony even when no single false item changes a policy decision.
How do Marcus, LeCun, Hinton, Bengio, and Bender differ?
They disagree less over whether AI systems have limitations than over which risks deserve the most attention, how likely future capability jumps are, and whether the right response is restraint, regulation, continued engineering, or a combination.
Do these 3 things before closing this tab:
1Fix the driver behind crashes, sound loss and screen glitches2Clear out junk files and repair common Windows errors3Scan for outdated or missing drivers - takes under a minute| Figure | Main criticism or concern | Risk emphasis | Preferred direction | Overlap with Marcus |
|---|---|---|---|---|
| Gary Marcus | Reliability, weak grounding, hype, and social deployment of current generative AI | Present misinformation, manipulation, intellectual-property concerns, and governance harms; future risks are possible but not needed to justify action | Stronger regulation and public oversight, plus research beyond simply scaling current LLMs | Baseline: useful bounded applications are possible, but current systems and deployment claims warrant skepticism |
| Yann LeCun | Overstated claims about current capability and existential danger | Current systems are limited; takeover narratives are remote or, in his view, preposterous | Continued engineering, open-source development, and safety guardrails rather than restricting progress to a few firms | Agrees that current systems have limitations and need safety work |
| Geoffrey Hinton | The trajectory toward systems that may exceed human intelligence | Future loss of control and the unprecedented challenge of more intelligent-than-human systems | Urgent safety research and public warning | Shares serious concern and distrust of complacency, but focuses more on a future control problem |
| Yoshua Bengio | Rapid capability gains alongside broad safety and societal risks | Present unreliability, democratic and national-security effects, and possible future control risks | Policy engagement and safety research alongside continued innovation | Shares caution about systems that can be highly capable yet unreliable |
| Emily Bender | Language-model hype, current harms, and policy distraction | Documented present harms rather than speculative rogue-AI scenarios | Evidence-led accountability and attention to labor, power, and social effects | Shares skepticism of hype and concern about present social damage |
The table shows why calling all five people simply “AI critics” can obscure more than it clarifies. Marcus and Bender begin with harms that can be observed now, but Marcus gives more attention to grounding, architecture and institutional regulation. Hinton gives the longest time horizon, asking what happens if systems become more capable than their designers. LeCun is also skeptical of exaggerated claims, but his preferred answer is more engineering and broad access. Bengio spans both periods, treating capability, reliability and public risk as parts of one policy problem.
How is Yann LeCun more optimistic than Marcus?
LeCun accepts that current LLMs are limited, but he treats those limitations more as engineering problems to improve than as grounds for broad restraint. Marcus emphasizes that weak grounding makes current deployment dangerous; LeCun emphasizes continued development, safety guardrails, and open access as a better route forward.
Rank #2
- 5-in-1 USB-C Hub: Experience comprehensive connectivity featuring a Power Delivery input, two USB-A 2.0 ports, a USB-A 3.0 port, and an HDMI port. (Note: The USB-C power delivery input port is only for connecting an external wall charger to power your laptop and cannot power peripheral devices.)
- 90W Pass-Through Charging: Achieve optimal charging with 90W pass-through power to your laptop, supported by a total input of 100W, with the hub reserving 10W for operational efficiency. (Note: Wall charger not included.)
- Quick Data Transfers: Accelerate your productivity with rapid data transfers using a high-speed 5Gbps USB 3.0 port and two 480Mbps USB 2.0 ports.
- 4K HDMI Display: Enhance your visual experience with a hub capable of delivering 4K resolution at 30Hz in both mirror and extend modes. Please note that this hub is compatible with MacBook (macOS 12 and newer), Windows 10 and 11, ChromeOS, and laptops equipped with DP Alt Mode and Power Delivery. Note: This device is not compatible with Linux.
- What You Get: Anker USB-C Hub (5-in-1, 4K HDMI), welcome guide, 18-month warranty, and our friendly customer service.
In a TIME interview published February 13, 2024, LeCun said that LLMs can “hallucinate” and “don’t really understand the real world,” while also calling existential-risk claims “preposterous.” He argues that progress with safeguards is preferable to restricting systems to a small number of firms. WIRED likewise describes him as rejecting dystopian scenarios and defending open-source AI.
The disagreement is not that LeCun thinks current models are flawless. Rather, Marcus sees limited grounding as a reason to slow or regulate risky deployment; LeCun sees it as a limitation that engineering can address, while opposing concentration of access. Their different forecasts about catastrophic risk shape how much weight each gives to openness versus precaution.
Recommended Free Tools
Open-source access is therefore a policy dividing line, not merely a technical preference. LeCun argues that keeping development broad can prevent a few companies from controlling the technology and its safeguards. Marcus’s emphasis is that wider deployment does not solve the underlying reliability problem; if systems remain easy to manipulate, openness can coexist with more opportunities for misuse. Both positions accept that guardrails matter, but they assign different risks to concentration and acceleration.
Why does Geoffrey Hinton focus more on future AI control?
Hinton’s warning centers on the possibility that AI systems may eventually become more intelligent than people and on humanity’s lack of experience controlling such entities. In a CBS News interview transcript from June 2024, he said, “for the first time ever we may have things more intelligent than us.”
That makes the contrast with Marcus partly a matter of time horizon. Marcus argues that today’s systems already warrant action because their unreliability and misuse can harm the information environment. Hinton gives greater weight to a future intelligence-and-control problem. These concerns are compatible: one can take present harms seriously without concluding that future superintelligence is imminent, and one can study future control risks without treating current harms as unimportant.
Hinton’s position also changes the standard for precaution. If a future system could outthink its operators, waiting for a clear real-world failure might be too late. Marcus does not need that forecast to justify intervention: the existing record of unreliable output, scalable deception and unlicensed use of creative work is sufficient, in his view, to demand oversight.
PC Slower Than It Used to Be?
A free scan shows the junk files, broken settings and background clutter dragging Windows down - then fixes them in one click.Free scan · Windows 10 & 11Outdated Drivers Are Slowing You Down
One free scan finds every outdated or missing driver and matches the right update for your exact hardware.Free scan · exact hardware matchWhere does Yoshua Bengio fit between current harms and future risk?
Bengio recognizes deep learning’s substantial capability gains while arguing that those gains bring a broad range of risks. In a July 25, 2023 post accompanying his Senate testimony, he wrote that the deep-learning revolution enables “tremendous progress and innovation” but “also entails a wide range of risks.”
His framing includes both present reliability concerns and wider policy questions. Bengio noted that GPT-4 could conduct highly proficient discussions that were sometimes unreliable. He also treats democratic effects and national security as serious considerations. Compared with Marcus, Bengio’s position is less centered on rejecting the current scaling trajectory and more on managing risks alongside continued innovation; both resist assuming that impressive capability guarantees dependable behavior.
Rank #3
- Sleek 7-in-1 USB-C Hub: Features an HDMI port, two USB-A 3.0 ports, and a USB-C data port, each providing 5Gbps transfer speeds. It also includes a USB-C PD input port for charging up to 100W and dual SD and TF card slots, all in a compact design.
- Flawless 4K@60Hz Video with HDMI: Delivers exceptional clarity and smoothness with its 4K@60Hz HDMI port, making it ideal for high-definition presentations and entertainment. (Note: Only the HDMI port supports video projection; the USB-C port is for data transfer only.)
- Double Up on Efficiency: The two USB-A 3.0 ports and a USB-C port support a fast 5Gbps data rate, significantly boosting your transfer speeds and improving productivity.
- Fast and Reliable 85W Charging: Offers high-capacity, speedy charging for laptops up to 85W, so you spend less time tethered to an outlet and more time being productive.
- What You Get: Anker USB-C Hub (7-in-1), welcome guide, 18-month warranty, and our friendly customer service.
Bengio’s middle position is important because it rejects two shortcuts. Capability gains are not proof that systems are safe, but evidence of risk is not proof that innovation should stop. Policy engagement and safety research are ways to address uncertainty while acknowledging that these systems can deliver valuable performance and still fail in consequential ways.
How does Emily Bender’s emphasis differ from Marcus’s?
Bender’s emphasis is on evidence of harms happening now and on preventing speculative rogue-AI scenarios from distracting policymakers. A Stanford Freeman Spogli Institute report attributes to her a warning that policymakers should not “fall for the distractions of AI hype.” The report describes her criticism as focused on present harms and treats rogue-AI narratives as speculative.
Marcus also challenges hype and stresses current social damage, but his critique reaches more directly into system reliability, the technical direction of LLM development, and the need for stronger institutions to regulate AI companies. In an April 2023 TED description, Marcus advocated a global nonprofit organization to regulate AI, warning that untrustworthy technology was being integrated at dangerously high speed. The difference is one of emphasis, not a simple divide between caring about current harms and ignoring them.
Bender’s evidence-led approach asks policymakers to examine who is harmed, whose labor and creative work are used, and how power is distributed. Marcus adds a forecast about the information environment: if deepfakes and fabricated references become cheap and abundant, democratic institutions may struggle to establish what is authentic. Their approaches overlap on accountability while differing in how much weight they give to technical architecture and future capability narratives.
Are present AI harms more important than future superintelligence?
There is no single answer shared by these critics, and the choice need not be either-or. Marcus’s case is that present harms already justify attention, even if AGI remains distant. Hinton’s concern is that a future system could exceed human intelligence and create an unprecedented control challenge. Bengio’s risk framing spans current unreliability, societal and security risks, and possible future dangers; Bender argues that speculative scenarios should not displace evidence about harms already occurring.
A useful way to read the disagreement is to separate four questions:
- What is happening now? Marcus, Bengio, and Bender all give weight to present problems, though they focus on different mechanisms and policy implications.
- How far can current approaches go? Marcus rejects claims that scaling current LLMs puts AGI near; LeCun emphasizes engineering progress while acknowledging limits.
- How should uncertainty be handled? Hinton presses the future control problem; Bender warns against letting speculative claims dominate policy; Marcus argues that concrete present risks are enough to support stronger oversight.
- What policies follow? Marcus favors stronger public oversight and regulation, LeCun favors open development with guardrails, and Bengio supports safety and policy engagement alongside innovation.
The practical distinction is that action on current harms does not require certainty about superintelligence. Equally, dismissing present models as imperfect does not resolve the separate question of what future capabilities might make possible.
What do their disagreements imply for regulation?
Marcus’s proposal for a global nonprofit regulatory organization reflects his concern that individual companies should not be the sole judges of systems they build and deploy. His April 2023 TED description presents rapid integration of untrustworthy technology as a governance problem requiring coordination beyond voluntary company assurances.
Rank #4
- Dual Converters, Infinite Potential：Includes 2× USB C male to USB A female adapters and 2× USB A male to USB C female adapters. Perfect for a wide range of uses—tablets with Bluetooth keyboards, expand USB ports on macbook, and more. Two different converters for all your daily needs
- Next-Level 10Gbps & 3A Charging: No more slow 480Mbps, this usb to usb c adapter has a transfer speed of up to 10Gbps, allowing you to do more transferring in less time. This usb adapter fits both USB A and USB C charger, supporting up to 3A fast charging
- Upgraded Exquisite Craftsmanship: With an aluminum alloy housing and metal connector, the usbc to usb adapter is extremely durable and sturdy. Rigorously tested to withstand more than 10,000 times of plugging and unplugging, ensuring long-lasting performance
- Broad Compatible: The usb c to usb adapter widely supports all USB C/ USB A devices like laptops, tablets, cellphones, car chargers, and phone chargers. Such as compatible with MacBook Pro/Air 2023/2022, Thunderbolt 4/3 Devices,Apple MagSafe Watch 9/8/7/SE/Ultra, iPad Pro 2022/2021, Samsung Galaxy S23/S20/S10, and iPhone 17/16/15 Pro. Plug and play
- Please Note: To reach 10Gbps speed, keep the cable under 3.3 ft. For USB A Male to USB C adapters, try flipping the USB C connector. USB C Male to USB A adapters support bidirectional 10Gbps transfer within 3.3 ft
LeCun’s preferred direction puts more weight on open development and safety guardrails, partly to avoid concentrating progress in a few firms. Bengio’s approach combines policy engagement with continued innovation and safety research. Bender’s emphasis would keep policymakers focused on documented effects, including labor, power and social harms, rather than allowing dramatic but unverified scenarios to set the entire agenda.
These are not interchangeable proposals. Regulation can target deployment and accountability; engineering can address technical limitations; public oversight can set boundaries that private incentives may not supply. The critics differ over which lever deserves priority, but none of the cited positions treats safety as irrelevant.
What’s actually slowing this PC down?
Pick the symptom - the matching free tool is one click away.
Is Gary Marcus anti-AI?
No. The position described in the IEEE Spectrum profile is critical of current generative AI’s reliability, hype, and deployment, but it allows for useful applications such as autocomplete and calls for AI that works better for people. His argument is for a different balance of research and accountability, not a ban on all AI.
Marcus’s proposed global nonprofit regulator, described by TED in April 2023, reflects his view that the pace of integration and the scale of potential harm require oversight beyond voluntary assurances from companies. It is a governance proposal tied to his concerns about trustworthiness, not evidence that he rejects every AI application.
What should readers take away from the disagreement?
“AI critic” does not describe one unified position. Marcus is distinctive for joining a critique of present reliability and social deployment to skepticism that scaling current LLMs is close to AGI. Hinton centers the possible future control problem; LeCun places more confidence in engineering and open-source progress; Bengio combines recognition of major capability gains with broad risk management; Bender prioritizes documented present harms and resistance to hype-driven policy distraction.
The disagreement is best understood by tracking the risk horizon, the expected trajectory of current architectures, the evidence standard for policy, and the preferred form of governance. Those axes explain why people can agree that models hallucinate or pose risks yet sharply disagree about what to do next.
For readers, the key distinction is between a forecast and a diagnosis. A forecast asks whether systems might become more intelligent than humans and difficult to control. A diagnosis asks what current systems do when they are unreliable, overtrusted or misused. Marcus keeps the diagnosis in the foreground; Hinton keeps the forecast in the foreground; Bengio considers both; LeCun expects engineering to narrow today’s limitations; and Bender insists that policy remain anchored to demonstrable effects.
Frequently Asked Questions
What does Gary Marcus mean when he calls generative AI “stupid” but dangerous?
He is contrasting fluent output with dependable grounding: a system may produce convincing material without reliably knowing whether it is true. That combination can invite overtrust and make manipulation more effective.
Best Value
- 5-in-1 Connectivity: Equipped with a 4K HDMI port, a 5 Gbps USB-C data port, two 5 Gbps USB-A ports, and a USB C 100W PD-IN port. Note: The USB C 100W PD-IN port supports only charging and does not support data transfer devices such as headphones or speakers.
- Powerful Pass-Through Charging: Supports up to 85W pass-through charging so you can power up your laptop while you use the hub. Note: Pass-through charging requires a charger (not included). Note: To achieve full power for iPad, we recommend using a 45W wall charger.
- Transfer Files in Seconds: Move files to and from your laptop at speeds of up to 5 Gbps via the USB-C and USB-A data ports. Note: The USB C 5Gbps Data port does not support video output.
- HD Display: Connect to the HDMI port to stream or mirror content to an external monitor in resolutions of up to 4K@30Hz. Note: The USB-C ports do not support video output.
- What You Get: Anker 332 USB-C Hub (5-in-1), welcome guide, our worry-free 18-month warranty, and friendly customer service.
Does Gary Marcus believe current large language models are close to AGI?
No. In the 2024 IEEE Spectrum profile, Marcus said he did not think the field was near AGI and described expectations of imminent AGI reflected in company valuations as unrealistic.
What book explains Gary Marcus’s views on AI regulation?
His book is Taming Silicon Valley: How We Can Ensure That AI Works for Us, published by MIT Press in 2023. The book addresses immediate risks including disinformation, deepfake pornography, and use of creative work without authorization.
What did Marcus warn about in his 2023 Calcalist interview?
On May 8, 2023, he warned that AI could contribute to a world where people trust nothing, complicating democratic life. He linked that concern to deepfakes and language models used to fabricate studies, data, and references.
How does Yann LeCun differ from Marcus on AI risk?
LeCun agrees that current LLMs hallucinate and do not understand the real world, but treats those limits as engineering problems. He rejects near-term extinction claims, favors open-source development with guardrails, and is more optimistic than Marcus about continued progress.
Why is Geoffrey Hinton more focused on future AI than Marcus?
Hinton emphasizes the possibility that machines could become more intelligent than humans, creating a control problem humanity has never faced. Marcus emphasizes harms from systems already deployed and does not require superintelligence to justify regulation.
What is Yoshua Bengio’s position on AI capability and safety?
Bengio recognizes tremendous progress from deep learning while warning of a wide range of risks. He notes that GPT-4 can be highly proficient yet unreliable and treats democratic, national-security and future-control questions as policy concerns.
Free tools Windows power users keep installed
One-click scans. No signup required.
What does Emily Bender mean by distractions from AI hype?
Bender’s emphasis is on documented present harms and on preventing speculative rogue-AI stories from displacing evidence-based policymaking. She overlaps with Marcus on skepticism toward hype but focuses less on architecture and global regulation.
Does Marcus support any uses of generative AI?
Yes. He identifies bounded applications such as autocomplete as useful. His criticism targets unreliable deployment, inflated AGI claims and weak accountability, not every AI application.
Why can present AI harms matter even if AGI is distant?
Unreliable systems can still generate convincing falsehoods, deepfakes and fabricated references at scale. Overtrust and misuse can damage information systems and democratic decision-making without any model being generally intelligent.
The Bottom Line
Gary Marcus’s distinctive claim is that present generative AI can be unreliable and dangerous at once. Hinton prioritizes the future control problem, LeCun trusts engineering and open access more, Bengio combines capability with broad risk management, and Bender centers documented harms. Their disagreement is mainly about time horizon, evidence and governance—not a simple choice between being for or against AI.
Quick wins for a faster PC:
Scan for outdated or missing drivers - takes under a minuteDriver Scan →Clear out junk files and repair common Windows errorsFree Scan →Quick Recap
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
