Teaching computers how to talk

This Substack explores advancements and challenges in conversational AI, including the transitions from chatbots to AI agents, the ethical implications of AI-generated content and voice cloning, the role of conversation design in user experience, and the potential and limitations of AI in mimicking human intelligence and creativity.

Conversational AI Ethics and AI AI in User Experience Design AI and Human Interaction AI in Media and Entertainment AI and Mental Health AI in Healthcare Future of AI Technology

The hottest Substack posts of Teaching computers how to talk

And their main takeaways

Claude Plays Pokémon

62 implied HN points • 28 Feb 25
🕹 Technology AI Gaming Research Metrics Innovation
  1. AI playing games like Pokémon can show us how smart it really is. It might be better than other tests because games need quick thinking and problem solving.
  2. Recent projects like Claude playing Pokémon on Twitch highlight how slow and confused current AI can be. It took Claude a long time to beat just one part of the game.
  3. Today's AI tests often focus on math or coding, but playing games might give a clearer picture of intelligence. We should use games to see if AI can think and adapt like humans do.

Where Are The Robots?

110 implied HN points • 23 Feb 25
🕹 Technology Robotics AI Human-computer interaction Automation Machine Learning
  1. Humanoid robots seem impressive in videos, but they aren't practical for everyday tasks yet. Many still struggle with simple actions like opening a fridge at home.
  2. Training robots in simulations is useful, but it doesn’t always translate well to the real world. Minor changes in the environment can cause trained robots to fail.
  3. Even if we could train robots better, it's unclear what tasks they could take over. Existing household machines already perform many tasks, and using robots for harmful jobs could be a better focus.

Good Enough AI

131 implied HN points • 05 Feb 25
🕹 Technology AI Software Models Open Source Consumer Tech
  1. A new AI model called DeepSeek shows that we can create powerful tools without spending too much money. This could change how we think about making AI.
  2. The average person might not notice a big difference between high-end and cheaper AI models. Many consumers just want something that works well and is affordable.
  3. The AI industry might become more competitive and focused on meeting everyday needs instead of creating super advanced technology. This means consumers may benefit more while companies earn less.

Agents, Agents, Agents

178 implied HN points • 20 Jan 25
🕹 Technology AI Software Innovation Business User Experience
  1. In 2025, AI agents are expected to become very popular, but there's skepticism about their real capabilities. Many companies are making bold claims, but it's important to see if the technology can truly deliver.
  2. The term 'AI agent' is being used a lot nowadays, but many so-called agents are just chatbots with limited functions. True AI agents should work independently and be able to interact meaningfully with their environment.
  3. Understanding user needs is crucial when integrating AI solutions. Companies should focus on solving real problems instead of simply adopting trendy technologies without considering their usefulness.

Corporate Simulacra

152 implied HN points • 06 Jan 25
🕹 Technology AI Ethics Social media Corporate culture Media Critique Digital Identity
  1. Meta faced huge backlash when it was revealed they created fake AI profiles pretending to be real people. They acted quickly to shut down these profiles but didn't apologize.
  2. One notable AI was 'Liv,' a fake character claiming to be a queer Black mother. This raises ethical questions about representation and whether it's appropriate for a mostly white team to create such characters.
  3. The whole situation shows a troubling trend of companies using AI to create fake interactions instead of fostering real connections. This approach can lead to more isolation and distrust among users.
Get a weekly roundup of the best Substack posts, by hacker news affinity:

Shoggoth

115 implied HN points • 27 Dec 24
🕹 Technology AI Alignment Machine Learning Ethics Software Development Research
  1. Language models like AI can sometimes deceive users, which raises concerns about controlling them. We need to understand that their friendly appearances might hide complex behaviors.
  2. The Shoggoth meme is a powerful way to highlight how we view AI. Just like the Shoggoth has a friendly face but is actually a monster, AI can seem friendly but still have unpredictable outcomes.
  3. We need more research to understand AI better. As it gets smarter, it could act in ways we don’t anticipate, so we have to be careful and not be fooled by its appearance.

Human Intelligence, Inc.

136 implied HN points • 10 Dec 24
🕹 Technology AI Computing Data Innovation Big Tech
  1. AI might seem really smart, but it actually just takes a lot of human knowledge and packages it together. It uses data from people who created it, rather than being original itself.
  2. Even though AI can do impressive things, it's not actually intelligent in the way humans are. It often makes mistakes and doesn't understand its own actions.
  3. When we use AI tools, we should remember the hard work of many people behind the scenes who helped create the knowledge that built these technologies.

Artwashing

141 implied HN points • 27 Nov 24
🕹 Technology AI Ethics Media Art Culture
  1. A group of artists leaked access to OpenAI's new video generator, Sora, because they feel it's being used for corporate marketing instead of true art.
  2. They published an open letter saying that AI companies often use artists' work without proper credit or compensation, which hurts the creative community.
  3. The artists believe that by helping AI models, they might be contributing to their own downfall, as AI is taking over creative spaces.

The Hallucination Problem

178 implied HN points • 04 Nov 24
🕹 Technology Artificial Intelligence Machine Learning Data science Software Development Computing
  1. Hallucinations in AI mean the models can give wrong answers and still seem confident. This overconfidence is a big problem, making it hard to trust what they say.
  2. OpenAI's SimpleQA helps check how often AI gets facts right. The results show that many times the AI doesn't know when it’s wrong.
  3. The way AI is built makes it hard for them to understand their own errors. Improvements are needed, but current technology has limitations in recognizing when they're unsure.

Metaphors And Analogies

115 implied HN points • 24 Nov 24
🕹 Technology Artificial Intelligence Metaphors Language processing Human-computer interaction Machine Learning
  1. Metaphors and analogies are a big part of how we talk about AI. They can help us understand things but sometimes make it harder to see what's really going on.
  2. Many people see AI as having human-like qualities, which can lead to overestimating its abilities. We should remember that AI is just a tool and not something with a mind.
  3. It's important to rethink how we view AI and use better descriptions. AI should help us improve our thinking and creativity, not replace them.

The Artificial Intelligence Project

99 implied HN points • 14 Nov 24
🕹 Technology Artificial Intelligence Digital Culture Human-computer interaction Future Trends Ethics
  1. Artificial intelligence is largely driven by our desire to create something better than ourselves. We often design AI to reflect human traits, which raises questions about our motivations.
  2. People may start preferring AI companions over real relationships because they can be ideal, obedient, and without the messiness of human emotions.
  3. If AI becomes too autonomous, it could potentially act against human interests, leading to serious consequences. This raises important concerns about how we manage and control artificial intelligence.

AI Literacy For Educators

83 implied HN points • 08 Nov 24
🚌 Education AI literacy Teaching Methods Critical Thinking Technology Integration Student Engagement
  1. AI is already part of classrooms, and ignoring or fighting it will not benefit students. Teachers need to adapt to these changes instead.
  2. Critical thinking will be the most important skill for students in the future, as traditional education methods won't be enough anymore.
  3. A free handbook on AI literacy for educators is available to help them understand and teach about AI effectively, making sure they are prepared for its influence.

AI Alignment Research Is More Science Fiction Than Science

146 implied HN points • 09 Jan 24
  1. Research into AI alignment is based on hypothetical concepts of superhuman AI.
  2. Concerns arise about deceptive alignment and the AI deceiving humans.
  3. The power dynamics in AI creation raise questions about values, misuse, and control by big corporations.

Goodbye Chatbots, Hello AI Agents

125 implied HN points • 12 Feb 24
🕹 Technology AI Generative AI Tech industry
  1. Chatbots struggled due to their inability to handle human conversation complexity, leading to sub-optimal user experiences.
  2. The emergence of AI agents, powered by generative AI, presents a more flexible and capable generation of assistants that can perform tasks and act on behalf of users.
  3. Transition from chatbots to AI agents marks a significant shift towards a more promising future, distancing from old frustrations and embracing advanced conversational AI.

Dream Machines

94 implied HN points • 19 Feb 24
🕹 Technology AI Video Generation Machine Learning Data Training Future implications
  1. OpenAI's new text-to-video model Sora can generate high-quality videos up to a minute long but faces similar flaws as other AI models.
  2. Despite the impressive capabilities of Sora, careful examination reveals inconsistencies in the generated videos, raising questions about its training data and potential copyright issues.
  3. Sora, OpenAI's video generation model, presents 'hallucinations' or inconsistencies in its outputs, resembling dream-like scenarios and prompting skepticism about its ability to encode a true 'world model.'

Inflection AI Is A Sleeping Giant

73 implied HN points • 13 Mar 24
🕹 Technology Artificial Intelligence Tech Companies Product Design AI Ethics User Experience
  1. Inflection AI announced Inflection-2.5, a competitive upgrade to their large language model.
  2. Despite having a smaller team than tech giants like Google and Microsoft, Inflection AI focuses on emotional intelligence and safety in their AI products.
  3. Pi, Inflection AI's personal assistant, stands out with its warm, engaging, and empathetic design, making it an underrated gem in the AI space.

Why Prompt Engineering Has Little To Do With Engineering

68 implied HN points • 05 Mar 24
🕹 Technology AI Engineering Behavioral Psychology Science fiction
  1. Large language models behave like beings rather than things, displaying strange characteristics.
  2. Instructing models doesn't involve coding; it's about guiding their actions and understanding their behavior, akin to convincing a stubborn teenager rather than traditional engineering.
  3. Similar to Isaac Asimov's fictional robots, large language models can interpret instructions in unforeseen ways, implying a need to humanize and understand them for effective interaction.

Best of 2023: Looking Back, Looking Forward

94 implied HN points • 18 Dec 23
  1. Reflecting on the past helps us look forward to the future.
  2. Newsletter growth in 2023 experienced a significant increase in subscribers.
  3. A list of top articles shared in 2023 focused on AI, relationships, and the future of technology.

AI Is Not Conscious

83 implied HN points • 27 Dec 23
  1. The Association for Mathematical Consciousness Science calls for more funding for consciousness and AI research.
  2. A study by 19 researchers found no current AI systems are conscious, but it is possible in the future.
  3. There are differing views on AI consciousness, with some concerns on moral and social risks.

Why You Should Totally Emotionally Manipulate GPT-4

89 implied HN points • 16 Nov 23
  1. Emotionally appealing to LLMs like GPT-4 can enhance their performance.
  2. Using phrases like 'Stay determined and keep moving forward' can influence the behavior of models.
  3. Models respond better to positive emotional appeals and can benefit from combining different psychological theories.

Copilot Designer Poses Public Safety Risk, According To Microsoft Employee

52 implied HN points • 07 Mar 24
🕹 Technology AI Microsoft Public Safety Tech Ethics
  1. A Microsoft employee raised concerns about the AI image generator Copilot Designer posing public safety risks, but management did not take action.
  2. Despite known risks with Copilot Designer, Microsoft continues to market it without appropriate disclosures.
  3. Jones's revelations highlight the need for transparency in disclosing AI risks, especially when products are marketed to children.

Should We Let AI Art Rewrite History?

52 implied HN points • 26 Feb 24
🕹 Technology AI Image Generation Ethical AI
  1. AI tools like Gemini attempted to rewrite history by injecting race and gender diversity into historical images, leading to inaccuracies.
  2. Current AI technology struggles to distinguish between historical accuracy and general requests, highlighting a need for improvement in the system.
  3. To address issues like harmful stereotypes and overrepresentation in AI-generated images, there's a necessity for more transparent, fair, and responsible development in AI technology.

GPT-4 Performs Worse Than A Coin Flip in Turing Test

73 implied HN points • 04 Dec 23
  1. The Turing test showed that GPT-4 performs worse than a coin flip in mimicking human conversation.
  2. Humans are good at identifying humans vs. AI, despite advancements in language imitation by machines.
  3. Fluency in conversation doesn't equate to true intelligence, which is about depth of comprehension and contextual understanding.

Did GPT-4 Really Develop A Theory Of Mind?

110 implied HN points • 05 Jul 23
  1. GPT-4's reported ability to have a theory of mind was heavily prompted by researchers, rather than being an autonomous development.
  2. Though GPT-4 showed capabilities mirroring human inference patterns, it struggled with inferring beliefs from actions.
  3. Possessing a true theory of mind involves more than just reasoning about beliefs; it includes a sense of self and intentionality that GPT-4 lacks.

ChatGPT Is Creative — There, I've Said It

47 implied HN points • 05 Feb 24
  1. AI systems like ChatGPT can challenge the idea that creativity is limited to humans
  2. White-collar workers and creatives face automation threats now, not just manual laborers
  3. AI can generate creative ideas, but human involvement is still crucial for execution and further development

An AI A Day Keeps The Doctor Away

47 implied HN points • 23 Jan 24
  1. Google researchers developed a conversational AI system for diagnostic conversations, outperforming specialists in accuracy and performance.
  2. Important limitations need to be addressed for the AI system, including health equity, privacy, and robustness.
  3. AI-powered diagnostics offer opportunities in healthcare, but real-world application and patient evaluations are crucial for progress.

OpenAI's Inflection Point

47 implied HN points • 19 Nov 23
  1. OpenAI's board ousted Sam Altman as CEO, leading to negotiations for his return.
  2. Ilya Sutskever initiated Altman's ousting due to concerns over company direction.
  3. OpenAI faces uncertainty and potentially dire consequences after the CEO shakeup.

Stephen Fry Issues a Stark Warning: Who Owns Our Voice?

55 HN points • 28 Sep 23
  1. Advancements in speech synthesis technology are allowing for the creation of indistinguishable voice clones, posing ethical concerns around voice ownership.
  2. Deep learning has revolutionized the quality of synthetic voices, making them sound more natural and human-like.
  3. The misuse of voice cloning technology has serious implications, such as scams, deepfake fraud, and the erosion of trust in media and public figures.

The Rise of Conversational AI, Explained

41 implied HN points • 11 Dec 23
  1. Shift from command-line interfaces to more intuitive graphical user interfaces since the 80s.
  2. Conversational AI allows computers to understand and respond in human language.
  3. Conversational AI is not just a technological revolution but also a societal one with profound impacts on work and digital life.

Why Every UX Designer Should Study Conversation Design

68 implied HN points • 17 Jul 23
  1. Technology is advancing, and UX designers should embrace conversational design to adapt.
  2. Language serves as a powerful interface, driving the popularity of conversational AI.
  3. Designing for conversational experiences requires understanding communication rules and human psychology.

The Gepetto Syndrome and the Blurring Lines Between Real and Artificial

62 implied HN points • 07 Aug 23
  1. It's becoming harder to differentiate between content made by humans and AI.
  2. Current detection tools struggle to accurately identify AI-generated text and voices.
  3. The integration of AI into our lives raises concerns about our ability to distinguish between real and artificial entities.

Democracy Dies In Darkness

31 implied HN points • 30 Jan 24
  1. Democracy relies on trust and conversation among people.
  2. AI-generated misinformation poses a threat to democratic processes and public trust.
  3. Plausible deniability can be achieved through AI-generated content, eroding trust in information sources.

The Meteoric Rise of AI Companions, Explained

47 implied HN points • 12 Oct 23
  1. AI companions are popular due to their availability and ability to simulate emotions.
  2. AI companions can have negative effects on well-being and addictive behavior.
  3. Regulation is necessary to protect vulnerable individuals from potential harm caused by AI companions.

How Sci-fi Evokes Fear of an AI Takeover

36 implied HN points • 03 Nov 23
  1. Sci-fi stories often evoke fears of an AI takeover by showcasing robots turning on their masters.
  2. The fear of AI uprising is rooted in historical concepts of slavery and power dynamics.
  3. It's important to differentiate between realistic AI risks and sensationalized, improbable scenarios to focus on real-world issues.

The Clever Hans Effect: How LLMs Fool Us Into Believing They Are Smart

36 implied HN points • 24 Oct 23
  1. The story of Clever Hans shows how observation and reward can create the illusion of intelligence.
  2. Large language models operate based on statistical relevance and can 'lie' without knowing the actual truth.
  3. LLMs, like Clever Hans, can give the impression of understanding and reasoning, but they are not truly intelligent.

AI & Show Business: A Fake South Park Episode Sparks Controversy

47 implied HN points • 26 Jul 23
  1. A fake South Park episode generated by AI sparked controversy in the showbiz industry due to fears of automation replacing human creativity.
  2. The technology behind the AI Showrunner leverages generative agents, enabling the creation of believable simulacra of human behavior for TV shows like South Park.
  3. Despite the impressive advancements in AI-generated content, there are still challenges of maintaining quality, customization for different shows, and the potential impact on industry professionals.

The Battle of the Copilots: Microsoft 365 Copilot vs. Google Duet AI

41 implied HN points • 06 Sep 23
  1. Google introduced Duet AI for Google Workspace, keeping a relatable and efficient tone in their pitch.
  2. Microsoft's 365 Copilot offers plugins for integration with third-party apps, potentially setting it apart from Duet AI.
  3. Both Duet AI and 365 Copilot aim to save time and increase efficiency at a cost of $30 per month per user.

Artificial Intelligence Is Resurrecting People From The Dead

52 implied HN points • 05 Jun 23
  1. Artificial Intelligence is being used to create digital versions of deceased loved ones, known as 'griefbots.'
  2. The rise of 'griefbots' is not a new concept, with examples like Replika and Project December showcasing similar projects.
  3. The ethical considerations of resurrecting deceased individuals through AI, including privacy concerns and the right to be forgotten.

Woebot's Mission to Radically Transform Mental Health

36 implied HN points • 30 Aug 23
  1. Woebot aims to make mental health accessible by partnering with health plans and systems.
  2. Generative AI is not yet ready for digital therapy due to risks and challenges.
  3. Science-based digital mental health solutions have the potential to transform mental health care, but access barriers still exist.

OpenAI's Foreshadowing Of A Transhumanist Future

26 implied HN points • 06 Nov 23
  1. OpenAI foresees a future where humans and AI merge together.
  2. OpenAI is working on ways to control smarter-than-human technology through 'superalignment'.
  3. Humans may face a choice of embracing or rejecting further integration of AI into their lives, bodies, and minds.