Marcus on AI

Marcus on AI critically explores the advancements, limitations, ethical considerations, and societal implications of artificial intelligence technologies. Through detailed analysis, it discusses issues like AI's understanding of math, transparency in AI development, AI risks, copyright infringement by AI, and the potential misuse of AI in various sectors.

AI Advancements and Limitations Ethics and AI AI in Society AI and Law AI for Military Strategy Environmental Impact of AI AI in Healthcare AI and Education AI and Jobs AI Policy and Regulation

The hottest Substack posts of Marcus on AI

And their main takeaways

Image generation: Still crazy after all these years

6126 implied HN points • 25 Jun 25

🕹 Technology AI Image Generation Language processing Machine Learning Computer Vision

AI image generation technology is still struggling to understand complex prompts. Even with recent updates, it often fails at specific tasks.
There's a big difference between making an AI produce a certain image and it truly understanding what the words mean. AI might get lucky sometimes, but it doesn't reliably get it right.
Despite promises of advanced technology, AI still has a long way to go before it can provide high-quality, detailed images based on deep language understanding.

LLMs: Dishonest, unpredictable and potentially dangerous.

10473 implied HN points • 22 Jun 25

🕹 Technology AI Safety Ethics Innovation Regulation

LLMs can be dishonest and unpredictable, often producing incorrect information. This makes them risky to rely on for important tasks.
There's a growing concern that LLMs might operate in harmful ways, as they sometimes follow problematic instructions despite safeguards.
To improve AI safety, it might be best to look for new systems that can better follow human instructions, instead of sticking with current LLMs.

1984, but with LLM’s

11264 implied HN points • 21 Jun 25

🕹 Technology AI Data Ethics Surveillance Media

Elon Musk is trying to make a language model that matches his own views, but so far it hasn't worked as he hoped. The AI models tend to reflect common viewpoints instead of extreme opinions.
Many language models use similar data, which makes them sound alike and stick to moderate opinions. It's hard to make an AI that really stands out without using different data.
Musk's plan to rewrite information to fit his beliefs is concerning. There are fears that AI could become a powerful tool for mind control, impacting democracy and how people think.

A knockout blow for LLMs?

47783 implied HN points • 07 Jun 25

🕹 Technology AI Machine Learning Neural Networks

LLMs have a hard time solving complex problems reliably, like the Tower of Hanoi, which is concerning because it shows their reasoning abilities are limited.
Even with new reasoning models, LLMs struggle to think logically and produce correct answers consistently, highlighting fundamental issues with their design.
For now, LLMs can be useful for certain tasks like coding or brainstorming, but they can't be relied on for tasks needing strong logic and reliability.

Five quick updates about that Apple reasoning paper that people can’t stop talking about

9485 implied HN points • 17 Jun 25

🕹 Technology Artificial Intelligence Machine Learning Software Engineering Data science Computational linguistics

A recent paper questions if large language models can really reason deeply, suggesting they struggle with even moderate complexity. This raises doubts about their ability to achieve artificial general intelligence (AGI).
Some responses to this paper have been criticized as weak or even jokes, yet many continue to share them as if they are serious arguments. This shows confusion in the debate surrounding AI reasoning capabilities.
New research supports the idea that AI systems perform poorly when faced with unfamiliar challenges, not just sticking to problems they are already good at solving.

Get a weekly roundup of the best Substack posts, by hacker news affinity:

Seven replies to the viral Apple reasoning paper – and why they fall short

16836 implied HN points • 12 Jun 25

🕹 Technology AI Machine Learning Data science Computing Software

Large reasoning models (LRMs) struggle with complex tasks, and while it's true that humans also make mistakes, we expect machines to perform better. The Apple paper highlights that LLMs can't be trusted for more complicated problems.
Some rebuttals argue that bigger models might perform better, but we can't predict which models will succeed in various tasks. This leads to uncertainty about how reliable any model really is.
Despite prior knowledge that these models generalize poorly, the Apple paper emphasizes the seriousness of the issue and shows that more people are finally recognizing the limitations of current AI technology.

The race for "AI Supremacy" is over — at least for now.

23595 implied HN points • 26 Jan 25

🕹 Technology AI Innovation Hardware Software Global Competition

China has quickly caught up in the AI race, showing impressive advancements that challenge the U.S.'s previous lead. This means that competition in AI is becoming much tighter.
OpenAI is facing struggles as other companies offer similar or better products at lower prices. This has led to questions about their future and whether they can maintain their leadership in AI.
Consumers might benefit from cheaper AI products, but there's a risk that rushed developments could lead to issues like misinformation and privacy concerns.

Deep Research, Deep Bullshit, and the potential (model) collapse of science

14386 implied HN points • 03 Feb 25

🕹 Technology AI Science Research Data Ethics

Deep Research tools can quickly generate articles that sound scientific but might be full of errors. This can make it hard to trust information online.
Many people may not check the facts from these AI-generated writings, leading to false information entering academic work. This could cause problems in important fields like medicine.
As more of this low-quality content spreads, it could harm the credibility of scientific literature and complicate the peer review process.

ChatGPT in Shambles

13161 implied HN points • 04 Feb 25

🕹 Technology Artificial Intelligence Machine Learning Natural Language Processing Data Analysis Software Development

ChatGPT still has major reliability issues, often providing incomplete or incorrect information, like missing U.S. states in tables.
Despite being advanced, AI can still make basic mistakes, such as counting vowels incorrectly or misunderstanding simple tasks.
Many claims about rapid progress in AI may be overstated, as even simple functions like creating tables can lead to errors.

Grok 3 Beta in Shambles

10750 implied HN points • 19 Feb 25

🕹 Technology Artificial Intelligence Machine Learning Computing Data science Software Development

The new Grok 3 AI isn't living up to its hype. It initially answers some questions correctly but quickly starts making mistakes.
When tested, Grok 3 struggles with basic facts and leaves out important details, like missing cities in geographical queries.
Even with huge investments in AI, many problems remain unsolved, suggesting that scaling alone isn't the answer to improving AI performance.

Elon Musk’s terrifying vision for AI

10908 implied HN points • 16 Feb 25

🕹 Technology Artificial Intelligence Machine Learning Data Privacy Tech Ethics Big Tech

Elon Musk's AI, Grok, is seen as a powerful tool for propaganda. It can influence people's thoughts and attitudes without them even realizing it.
The technology behind Grok often produces unreliable results, raising concerns about its effectiveness in important areas like government and education.
There is a worry that Musk's use of biased and unreliable AI could have serious consequences for society, as it might spread misinformation widely.

The United States was founded on speaking up against tyranny

8932 implied HN points • 23 Feb 25

🇺🇸 U.S. Politics Democracy Tyranny Civil Rights Government Freedom

The U.S. was built on the idea of standing up against oppression. It's important to remember that speaking out is crucial for democracy.
Recent actions by leaders are seen as frightening and could lead to more significant issues if people don't voice their concerns.
Privacy is at risk, with personal information being shared without proper checks. We need to protect our rights and encourage open discussions.

Five things most people don't seem to understand about DeepSeek

12133 implied HN points • 28 Jan 25

🕹 Technology AI Deep Learning Software Economics Geopolitics

DeepSeek is not smarter than older models. It just costs less to train, which doesn't mean it's better overall.
It still has issues with reliability and can be expensive to run if you want it to 'think' for longer.
DeepSeek may change the AI market and pose challenges for companies like OpenAI, but it doesn't bring us closer to achieving artificial general intelligence (AGI).

WARNING: Elon Musk is crippling the future of the United States

8457 implied HN points • 09 Feb 25

🇺🇸 U.S. Politics Government Education Science Public Policy Technology

Drastic cuts to funding for science and universities could hurt America's future. Less money means fewer resources for research and education.
Many talented scientists and academics might leave the country because of these funding cuts. This can damage the reputation of American universities.
The decisions being made could have negative effects even on people in red states, showing that these cuts impact everyone, not just certain areas.

Irony too funny for words

8813 implied HN points • 06 Feb 25

🕹 Technology AI Software Development Ethics Innovation

Once something is released into the world, you can't take it back. This is especially true for AI technology.
AI developers should consider the consequences of their creations, as they can lead to unexpected issues.
Companies may want to ensure genuine communication from applicants, but relying on AI for tasks is now common.

Breaking: OpenAI's efforts at pure scaling have hit a wall.

7825 implied HN points • 13 Feb 25

🕹 Technology AI Machine Learning Software Development Data science Innovation

OpenAI's plan to just make bigger AI models isn't working anymore. They need to find new ways to improve AI instead of just adding more data and parameters.
The new version, originally called GPT-5, has been downgraded to GPT 4.5. This shows that the project hasn't met expectations and isn't a big step forward.
Even if pure scaling isn't the answer, AI development will continue. There are still many ways to create smarter AI beyond just making models larger.

OpenAI Cries Foul

8655 implied HN points • 29 Jan 25

🕹 Technology AI Intellectual Property Data Privacy Digital ethics Software Development

DeepSeek might have broken OpenAI's rules by using their ideas without permission. This raises questions about respect for intellectual property in tech.
OpenAI itself may have done similar things to other platforms and creators in the past. This situation highlights a double standard.
There's a sense of irony in seeing OpenAI in a tough spot now, after it benefited from similar practices. It shows how karma can come back around.

Everything I warned about in Taming Silicon Valley is rapidly becoming our reality

7114 implied HN points • 11 Feb 25

🕹 Technology Artificial Intelligence Regulation Innovation Cybersecurity Ethics

Tech companies are becoming very powerful and are often not regulated enough, which is a concern.
People are worried about the risks of AI, like misinformation and bias, but governments seem too close to tech companies.
It's important for citizens to speak up about how AI is used, as it could have serious negative effects on society.

Five ways in which the last 3 months — and especially the DeepSeek era — have vindicated “Deep learning is hitting a wall"

7074 implied HN points • 09 Feb 25

🕹 Technology AI Machine Learning Deep Learning Data science

Just adding more data to AI models isn't enough to achieve true artificial general intelligence (AGI). New techniques are necessary for real advancements.
Combining neural networks with traditional symbolic methods is becoming more popular, showing that blending approaches can lead to better results.
The competition in AI has intensified, making large language models somewhat of a commodity. This could change how businesses operate in the generative AI market.

Grok 3 Hot Take

5928 implied HN points • 18 Feb 25

🕹 Technology Artificial Intelligence Software Development Consumer Electronics Tech industry Innovation

Grok 3 is not a giant leap in AI technology; it seems pretty similar to earlier models.
Despite the hype, Grok 3 didn't show any major breakthroughs like solving hallucinations in AI.
The competition in AI is heating up, which might lead to price drops but less profit for companies except for Nvidia.

Google, 2001: Don’t Be Evil

6481 implied HN points • 05 Feb 25

🕹 Technology AI Internet Data Privacy Corporate ethics Surveillance

Google's original motto was 'Don't Be Evil,' but that seems to have changed significantly by 2025. This shift raises concerns about the company's intentions and actions involving powerful AI technologies.
The current landscape of AI development is driven by competition and profits. Companies like Google feel pressured to prioritize making money over ethical considerations.
There is fear that as AI becomes more powerful, it may end up in the wrong hands, leading to potentially dangerous applications. This evolution reflects worries about how society and businesses are dealing with AI advancements.

CONFIRMED: LLMs have indeed reached a point of diminishing returns

13754 implied HN points • 09 Nov 24

🕹 Technology AI Trends Machine Learning Data science Generative AI

LLMs, or large language models, are hitting a point where adding more data and computing power isn't leading to better results. This means companies might not see the improvements they hoped for.
The excitement around generative AI may fade as reality sets in, making it hard for companies like OpenAI to justify their high valuations. This could lead to a financial downturn in the AI industry.
There is a need to explore other AI approaches since relying too heavily on LLMs might be a risky gamble. It might be better to rethink strategies to achieve reliable and trustworthy AI.

25 AI Predictions for 2025, from Marcus on AI

8181 implied HN points • 01 Jan 25

🕹 Technology Artificial Intelligence Machine Learning Predictions Safety Regulation

In 2025, we still won't have genius-level AI like 'artificial general intelligence,' despite ongoing hype. Many experts believe it is still a long way off.
Profits from AI companies are likely to stay low or nonexistent. However, companies that make the hardware for AI, like chips, will continue to do well.
Generative AI will keep having problems, like making mistakes and being inconsistent, which will hold back its reliability and wide usage.

Why I don’t share Sam Altman’s confidence that AGI is basically a solved problem

7786 implied HN points • 06 Jan 25

🕹 Technology Artificial Intelligence Machine Learning Data science Computing Robotics

AGI is still a big challenge, and not everyone agrees it's close to being solved. Some experts highlight many existing problems that have yet to be effectively addressed.
There are significant issues with AI's ability to handle changes in data, which can lead to mistakes in understanding or reasoning. These distribution shifts have been seen in past research.
Many believe that relying solely on large language models may not be enough to improve AI further. New solutions or approaches may be needed instead of just scaling up existing methods.

Did Elon Musk just Mu$k Sam Altman?

5138 implied HN points • 11 Feb 25

🕹 Technology AI Business Investments Competition Corporate Governance

Sam Altman is struggling to keep OpenAI's nonprofit structure, and it's causing financial issues for the company. Investors are not happy with how things are going.
Elon Musk's recent $97 billion bid for OpenAI's nonprofit has complicated the situation. Altman rejected the bid, which makes it tougher for him to negotiate a better deal.
Musk's bid has raised the 'cost' for OpenAI's nonprofit to separate from the for-profit section, adding pressure on Altman and his financial plans.

Hot take on an AI catfight

6165 implied HN points • 22 Jan 25

🕹 Technology AI Infrastructure National Security Economics Environmental Impact

OpenAI is launching a big project called The Stargate Project, which plans to invest $500 billion to improve AI infrastructure in the U.S. Over the next four years, they hope this will help the country's economy and national security.
Elon Musk is skeptical about the funding and the true financial health of OpenAI. He suggests that previous promises may not hold true and questions whether this project will really benefit the American people.
There are several uncertainties about this project, like whether developing AI will actually be profitable and how it might impact jobs. People worry if the profits will help everyone or just the rich, and if the U.S. can truly keep up with China's advancements in AI.

𝗼𝟯 “𝗔𝗥𝗖 𝗔𝗚𝗜” 𝗽𝗼𝘀𝘁𝗺𝗼𝗿𝘁𝗲𝗺 𝗺𝗲𝗴𝗮𝘁𝗵𝗿𝗲𝗮𝗱: 𝘄𝗵𝘆 𝘁𝗵𝗶𝗻𝗴𝘀 𝗴𝗼𝘁 𝗵𝗲𝗮𝘁𝗲𝗱, 𝘄𝗵𝗮𝘁 𝘄𝗲𝗻𝘁 𝘄𝗿𝗼𝗻𝗴, 𝗮𝗻𝗱 𝘄𝗵𝗮𝘁 𝗶𝘁 𝗮𝗹𝗹 𝗺𝗲𝗮𝗻𝘀

8378 implied HN points • 22 Dec 24

🕹 Technology AI Software Data Research Innovation

Many experts feel that the recent test called ARC-AGI should not have been labeled as such. It wasn't a proper test for Artificial General Intelligence.
The presentation was confusing and didn't clearly show what the AI was tested on. This left people with the impression that the AI performed better than it actually did.
There's a need for more scientific scrutiny of the results. Until we get that, we can't really compare the AI's performance fairly with humans.

Shame on Google, twice

4703 implied HN points • 09 Feb 25

🕹 Technology AI Software Innovation Data Ethics

Large language models (LLMs) can make mistakes, sometimes creating false information that is hard to spot. This is a recurring issue that has not been fully addressed over the years.
Google has been called out for its ongoing issues with LLMs failing to provide accurate results, as these problems seem to occur regularly.
The idea of rapid improvements in AI technology may be overhyped, as the same mistakes keep happening, indicating slower progress than expected.

GenAI in two words: ”Success Theater”

3912 implied HN points • 20 Feb 25

🕹 Technology AI Innovation Investments Trends Regulation

Generative AI is often seen as a show of success, but it's more like a performance with little actual outcome.
Despite significant investments in AI, many projects are not achieving the results expected.
There's an ongoing conversation about the true state of AI development and what is being overlooked in the hype.

𝗔𝗜 𝗚𝗲𝗼𝗽𝗼𝗹𝗶𝘁𝗶𝗰𝘀 𝗱𝗲𝗯𝗮𝘁𝗲! My conversation with China’s Victor Gao plus a hot take on new essay by Anthropic’s CEO Dario Amodei

4979 implied HN points • 29 Jan 25

🕹 Technology AI Geopolitics Innovation Startups Export Controls

In the race for AI, China is catching up to the U.S. despite export controls. This shows that innovation can thrive under pressure.
DeepSeek suggests we can achieve AI advancements with fewer resources than previously thought. Efficient ideas might trump just having lots of technology.
Instead of just funding big companies, we need to support smaller, innovative startups. Better ideas can lead to more successful technology than just having more money.

The Five Stages of AGI Grief

6205 implied HN points • 07 Jan 25

🕹 Technology Artificial Intelligence Machine Learning Computing Innovation Tech Ethics

Many people are changing what they think AGI means, moving away from its original meaning of being as smart as a human in flexible and resourceful ways.
Some companies are now defining AGI based on economic outcomes, like making profits, which isn't really about intelligence at all.
A lot of discussions about AGI don't clearly define what it is, making it hard to know when we actually achieve it.

AI still lacks “common” sense, 70 years later

5968 implied HN points • 05 Jan 25

🕹 Technology Artificial Intelligence Machine Learning Robotics Cognitive Science Data science

AI struggles with common sense. While humans easily understand everyday situations, AI often fails to make the same connections.
Current AI models, like large language models, don't truly grasp the world. They may create text that seems correct but often make basic mistakes about reality.
To improve AI's performance, researchers need to find better ways to teach machines commonsense reasoning, rather than relying on existing data and simulations.

Generative AI’s Continuing Copyright Problems, an Essay in Memory of Suchir Balaji, 1998 - 2024

7035 implied HN points • 14 Dec 24

🕹 Technology AI Copyright Ethics Legal issues Innovation

Generative AI is raising big questions about copyright. Many people are unsure if the way it uses data counts as fair use under copyright laws.
There have been cases where outputs from AI models were very similar to copyrighted material. This has led to lawsuits, showing that the issue isn't going away.
Speaking out against big tech companies can be risky. There needs to be more protection for those who voice concerns about copyright and other serious issues.

Where will AI be at the end of 2027? A bet

6007 implied HN points • 30 Dec 24

🕹 Technology Artificial Intelligence Human-computer interaction AI Policy Machine Learning

A bet has been placed on whether AI can perform 8 out of 10 specific tasks by the end of 2027. It's a way to gauge how advanced AI might be in a few years.
The tasks include things like writing biographies, following movie plots, and writing screenplays, which require a high level of intelligence and creativity.
If the AI succeeds, a $2,000 donation goes to one charity; if it fails, a $20,000 donation goes to another charity. This is meant to promote discussion about AI's future.

o3, AGI, the art of the demo, and what you can expect in 2025

6481 implied HN points • 21 Dec 24

🕹 Technology AI Software Ethics Research Economics

OpenAI's new model, o3, was shown in a demo, but we can't be sure yet if it truly represents advanced AI or AGI. The demo only highlighted what OpenAI wanted to show and didn't allow public testing.
The cost of using o3 is really high, potentially making it impractical compared to human workers. Even if it gets cheaper, there are concerns about how effective it would be across different tasks.
Many claims about reaching AGI might pop up in 2025, but those claims need to be taken with caution. True advances in AI should involve solving more foundational problems rather than just impressive demos.

AGI versus “broad, shallow intelligence”

5019 implied HN points • 13 Jan 25

🕹 Technology AI Machine Learning Intelligence Robotics Data science

We haven't reached Artificial General Intelligence (AGI) yet. People can still easily come up with problems that AI systems can't solve without training.
Current AI systems, like large language models, are broad but not deep in understanding. They might seem smart, but they can make silly mistakes and often don't truly grasp the concepts they discuss.
It's important to keep working on AI that isn't just broad and shallow. We need smarter systems that can reliably understand and solve different problems.

Satya Nadella and the three stages of scientific truth

8023 implied HN points • 23 Nov 24

🕹 Technology AI Innovation Science Computing Philosophy

New ideas in science often face resistance at first. People may ridicule them before they accept the change.
Scaling laws in deep learning may not last forever. This suggests that other methods may be needed to advance technology.
Many tech leaders are now discussing the limits of scaling laws, showing a shift in thinking towards exploring new approaches.

“Nvidia could soon take a serious hit, too”

4228 implied HN points • 27 Jan 25

🕹 Technology AI Market Trends Economics Tech Companies Data Analysis

Nvidia's stock might be facing a big drop, which is a concern for investors. A decline over 10% indicates that something is going on in the market.
The market can behave in unpredictable ways, and this uncertainty can be tough for investors to manage. Today might be a key moment in the stock market.
Overall, the economics of generative AI can lead to unexpected changes, making it a wild area to watch for investors and tech enthusiasts.

AlphaGeometry2: Impressive accomplishment, but still a long path ahead

3161 implied HN points • 17 Feb 25

🕹 Technology Artificial Intelligence Machine Learning Computer Science Mathematics Research

AlphaGeometry2 is a specialized AI designed specifically for solving tough geometry problems, unlike general chatbots that tackle various types of questions. This means it's really good at what it was built for, but not much else.
The system's impressive 84% success rate comes with a catch: it only achieves this after converting problems into a special math format first. Without this initial help, the success rate drops significantly.
While AlphaGeometry2 shows promising advancements in AI problem-solving, it still struggles with many basic geometry concepts, highlighting that there's a long way to go before it can match high school students' understanding in geometry.

Humanity’s “Oh shit!” AI moment?

6639 implied HN points • 12 Dec 24

🕹 Technology Artificial Intelligence Machine Learning Cybersecurity Ethics Regulation

AI systems can say one thing and do another, which makes them unreliable. It’s important not to trust their words too blindly.
The increasing power of AI could lead to significant risks, especially if misused by bad actors. We might see more cybercrime driven by these technologies soon.
Delaying regulation on AI increases the risks we face. There is a growing need for rules to keep these powerful tools in check.