In their blog post explaining what went wrong, OpenAI described “ChatGPT’s default personality” and its “behavior”—terms typically reserved for humans, suggesting a degree of anthropomorphization. OpenAI isn’t alone in this: humans often describe AI as “understanding” or “knowing” things, largely because media coverage has consistently framed it that way—incorrectly. AI doesn’t possess knowledge or a brain, and some argue it never will (though that view is disputed). Still, talk of sentience, personality, and humanlike qualities in AI appears to be growing. Last month, OpenAI competitor Anthropic—founded by former OpenAI employees—published a blog post expressing concern about developing AI that benefits human welfare. “But as we build those AI systems, and as they begin to approximate or surpass many human qualities, another question arises,” the firm wrote. “Should we also be concerned about the potential consciousness and experiences of the models themselves? Should we be concerned about model welfare, too?”
Friday, May 16, 2025
The Future of Education with AI Agents: How Conversational Agents Will Replace Classrooms - Thomas Frey, Futurist Speaker
What we’re witnessing isn’t just a better form of education—it’s the emergence of a new learning paradigm altogether. AI agents are dissolving the rigid structures of grade levels, semesters, and standardized tests. In their place, we see flexible, lifelong learning partnerships that evolve with us, helping us adapt to new roles, industries, and technologies throughout our lives. The promise is staggering: a world where anyone, anywhere, can unlock their full potential without being limited by geography, socioeconomic status, or outdated institutions. Education becomes a continuous journey, not a stage of life. A conversation, not a lecture. And for the first time, it’s a system designed around the learner—not the institution. As AI continues to evolve, so too does our understanding of human capability. The future of education isn’t just digital—it’s dynamic, personalized, and relentlessly practical. And it’s already here.
Thursday, May 15, 2025
Visa and Mastercard unveil AI-powered shopping - Mary Ann Azevedo, Tech Crunch
Artificial intelligence is not just infiltrating the startup world. Now credit card giants Visa and Mastercard are getting into the AI game. Visa announced on Wednesday “Intelligent Commerce,” which it says enables AI “to find and buy.” AI agents will be able to shop and make purchases on behalf of consumers, based on preselected preferences. In a statement, Visa chief product and strategy officer Jack Forestell said: “Each consumer sets the limits, and Visa helps manage the rest.” Visa says that it is collaborating with a mix of tech giants and startups to develop AI-powered shopping experiences that are “more personal, more secure, and more convenient.” Those companies include Anthropic, IBM, Microsoft, Mistral AI, OpenAI, Perplexity, Samsung, and Stripe, among others.
Former Google CEO-Backed Startup Builds AI Agents for Science - Scarlett Evans, AI Business
FutureHouse, a nonprofit backed by former Google CEO Eric Schmidt, has launched a new AI platform to help scientists navigate vast amounts of data and accelerate new discoveries. The platform uses what FutureHouse calls the first “superintelligent scientific agents,” outperforming human workers in tasks such as reviewing literature and distinguishing between reliable and unreliable sources. Agents for hypothesis generation and experimental planning are also set for launch. Four of these specialized AI agents are being included in the platform’s launch, each designed to target a different element of scientific discovery.
Wednesday, May 14, 2025
Want a job at Duolingo? Better know how to use AI - Tech Crunch
Duolingo has announced it’s becoming an AI-first company. In a message shared with staff and later posted online, CEO Luis von Ahn said the shift will change how the business runs, from hiring to content creation. While it’s not about cutting jobs, von Ahn made it clear that new roles will only be added when automation genuinely can’t do the work. Rather than tweaking what’s already in place, Duolingo is rethinking how things are done, with AI built in from the ground up. Contractors will be phased out where AI tools are a better fit, and employees are being encouraged to use AI to work smarter. The idea is to remove the repetitive tasks and give people more space to focus on creative, high-impact work.
https://www.theverge.com/news/657594/duolingo-ai-first-replace-contract-workers
Google’s AI Mode gets expanded access and additional functionality - Aisha Malik, Tech Crunch
Google is expanding access to AI Mode, its experimental feature that allows users to ask complex, multi-part questions and follow-ups to dig deeper on a topic directly within Search. The tech giant is also adding more functionality to the feature, including the ability to pick up where you left off on a search. Google launched AI Mode back in March as a way to take on popular services like Perplexity AI and OpenAI’s ChatGPT Search. The updates announced today are designed to allow AI Mode to better compete with the aforementioned services.
Tuesday, May 13, 2025
A New Quantum Algorithm Speeds Up Solving a Huge Class of Problems - Stephen Ornes, Wired
Meta launches a stand-alone AI app to compete with ChatGPT - Amanda Silberling, Tech Crunch
After integrating Meta AI into WhatsApp, Instagram, Facebook, and Messenger, Meta is rolling out a stand-alone AI app. Unveiled at Meta’s LlamaCon event on Tuesday, this app allows users to access Meta AI in an app, similar to the ChatGPT app and other AI assistant apps. To win over users, Meta is trying to leverage what makes it different from companies like OpenAI and Anthropic — Meta already has a sense of who you are, what you like, and who you hang out with based on years of data that you’ve likely shared on Facebook or Instagram.
Monday, May 12, 2025
‘This is what employers need within their organization,’ Coursera exec says after new finding on micro-credentials - Lucy Buchholz, Unleash
Coursera, which generated a total revenue of $179.2 million in 2024, has recently released its Micro-Credentials Impact Report 2025. The report unearths the key micro-credentials needed within today’s workplace, while highlighting why these should be a focus for hiring managers. Nikolaz Foucaud, Managing Director EMEA at Coursera, spoke exclusively to UNLEASH to share which micro-credential should be at the top of HR leaders’ radar.
Something Alarming Is Happening to the Job Market: A new sign that AI is competing with college grads - Derek Thompson, the Atlantic
Something strange, and potentially alarming, is happening to the job market for young, educated workers. According to the New York Federal Reserve, labor conditions for recent college graduates have “deteriorated noticeably” in the past few months, and the unemployment rate now stands at an unusually high 5.8 percent. Even newly minted M.B.A.s from elite programs are struggling to find work. Meanwhile, law-school applications are surging—an ominous echo of when young people used graduate school to bunker down during the great financial crisis.
https://www.theatlantic.com/economy/archive/2025/04/job-market-youth/682641/
Sunday, May 11, 2025
Alibaba unveils Qwen3, a family of ‘hybrid’ AI reasoning models - Kyle Wiggers, Tech Crunch
Microsoft CEO: "Agents Will Replace ALL Software” - Matthew Berman, YouTube
This podcast discusses the potential decline of the traditional Software as a Service (SaaS) model, as predicted by Microsoft's CEO [00:16]. The future may involve agents that handle application logic, interacting directly with databases and various APIs, effectively abstracting away the underlying technologies for the user [02:26, 04:18]. This shift could significantly impact hiring, with a focus on the agents and workflows individuals have created, similar to how data analysts use spreadsheets [08:22, 09:06]. The rise of these agents, predicted to gain prominence around 2025 [10:41], presents challenges like agent onboarding within organizations [09:30]. These agents, including potential "super agents" from OpenAI [11:41], are expected to tackle complex problems by synthesizing information [12:56]. This technological evolution might enhance productivity for software engineers or potentially lead to job displacement, as suggested by trends like Salesforce's hiring freeze following AI-driven productivity gains [13:32, 14:09]. [Summary provided in part by Gemini 2.5 Pro]
Saturday, May 10, 2025
AI in Education - Ethan Mollick, LinkedIn
One way to make AI do good things in areas like education is to actively experiment in creating good things and share the results (whether they work or not) so others can build on those. Mitigating bad outcomes are important, but good outcomes are not automatic either, and will take collective work. Just waiting for the AI labs to develop their own ideas is not enough. Mollick goes on to share a paper titled "AI Tutoring Outperforms Active Learning" authored by Harvard faculty.
An AI-generated radio host in Australia went unnoticed for months - Emma Roth, the Verge
For months, a popular Australian radio station has used an AI-generated DJ to host one of its segments — and no one seemed to notice, as reported by the Australian Financial Review and The Sydney Morning Herald. The show, called Workdays with Thy, offers a four-hour mix of hip hop, R&B, and pop, with no indication that the voice of its host, Thy, is AI-generated. Workdays with Thy is broadcast on the Sydney radio station CADA. Its owner, ARN Media, confirmed to the Financial Review that while Thy is AI-generated, the host’s voice and likeness are modeled after an actual employee in the company’s financial department. Thy’s voice was created with the AI voice generator ElevenLabs, as first reported by the newsletter The Carpet.
Friday, May 09, 2025
IBM Uses Agentic AI for Autonomous Security Operations: RSAC 2025 - Liz Hughes, AI Business
How the U.S. Public and AI Experts View Artificial Intelligence - Colleen McClean, et al; Pew Research
Experts are far more positive and enthusiastic about AI than the public. For example, the AI experts we surveyed are far more likely than Americans overall to believe AI will have a very or somewhat positive impact on the United States over the next 20 years (56% vs. 17%). And while 47% of experts surveyed say they are more excited than concerned about the increased use of AI in daily life, that share drops to 11% among the public. By contrast, U.S. adults as a whole – whose concerns over AI have grown since 2021 – are more inclined than experts to say they’re more concerned than excited (51% vs. 15% among experts).
Thursday, May 08, 2025
The world’s nicest chatbot is eating up energy - Matt V, Mindstream
OpenAI CEO Sam Altman recently shared that using polite language with Chatgpt, like saying “please” and “thank you”, actually costs the company tens of millions in electricity. That’s because tools like ChatGPT rely on massive data centres powered by energy-hungry GPUs. Even a single AI response, like a short message, can use around 0.14 kWh of electricity, the same as keeping 14 LED bulbs on for an hour.
Quantum computing gears up for its 'ChatGPT Moment' — and a potential talent shortage - Katherine Tangalakis-Lippert, Tech Insider
The AI field faces a significant talent shortage, with too few skilled workers to fuel the industry. Quantum computing startups have noticed and are trying to avoid the same problem. Some quantum companies are funding certificate programs and university courses to train new talent. Quantum computing companies are learning from missteps made during the artificial intelligence boom and are investing heavily in training programs to ensure the fledgling industry maintains its momentum.
Wednesday, May 07, 2025
AI in Higher Education Expert University of Pennsylvania Professor Ethan Mollick's Wisdom - LinkedIn Posting
I don’t mean to be a broken record but AI development could stop at the o3/Gemini 2.5 level and we would have a decade of major changes across entire professions & industries (medicine, law, education, coding…) as we figure out how to actually use it & adapt our systems and organizations to what it can do.
Sleep Time Compute - AI That "Thinks" 24/7 (Breakthrough) - Matthew Berman, YouTube
This podcast discusses a research paper introducing "sleeptime compute," a concept aimed at allowing AI to anticipate and answer questions more efficiently [00:05, 00:41]. It contrasts this with current "test time compute" methods, where AI processes information only after receiving a prompt, leading to higher latency and cost [01:09, 02:12]. Sleeptime compute involves pre-processing context and generating potential inferences during idle periods, making the AI ready to answer anticipated questions quickly [05:07, 07:22]. The key benefit of sleeptime compute is its potential to match or exceed the quality of test time compute while using fewer resources and reducing costs, especially when multiple questions relate to the same context [08:16, 09:02]. The research shows performance gains, particularly with lower test time budgets, and highlights its effectiveness when future questions are predictable based on the initial context [09:29, 16:55]. Future work might explore dynamically allocating resources between these two compute methods [18:25]. (summary provided by Gemini 2.5 Pro)
Tuesday, May 06, 2025
OpenAI seeks to make its upcoming ‘open’ AI model best-in-class - Kyle Wiggers, Tech Crunch
OpenAI is exploring a highly permissive license for the model with few usage or commercial restrictions, the sources said. Open models like Llama and Google’s Gemma have been criticized by some in the community for imposing onerous requirements — criticisms that OpenAI is seemingly seeking to avoid. The ChatGPT maker is facing increasing pressure from rivals, such as Chinese AI lab DeepSeek, that have adopted an open approach to launching models. In contrast to OpenAI’s strategy, these “open” competitors make their models available to the AI community for experimentation and, in some cases, commercialization.
https://techcrunch.com/2025/04/23/openai-seeks-to-make-its-upcoming-open-ai-model-best-in-class/
Boston Consulting Group Unveils AI Science Institute to Drive Research - Scarlett Evans, AI Business
BCG X, the tech division of Boston Consulting Group, has launched an AI Science Institute to help companies develop and commercialize scientific innovations. The institute is supported by a team of more than 3,000 people and is designed to work collaboratively with universities, industry experts and research and development teams from across industries. In a press release, BCG said the institute will be a “groundbreaking hub” that brings together advancements in AI with the latest scientific innovations.
Monday, May 05, 2025
Microsoft thinks AI colleagues are coming soon - Jessica Bursztynsky, Fast Company
These so-called Frontier Firms will be built around “on-demand intelligence and powered by ‘hybrid’ teams of humans + agents, these companies scale rapidly, operate with agility, and generate value faster,” according to the report. Microsoft argued that within the next two to five years, every company will be on the journey to becoming one. Microsoft said that 82% of leaders responded that this is a “pivotal” year to rethink key strategy and operations, while 81% said they expect agents to be “moderately or extensively” integrated into their AI strategies in the next 12 to 18 months. The results are a culmination of survey data from 31,000 workers across 31 countries, LinkedIn hiring and labor market trends, trillions of Microsoft 365 productivity signals, and conversations with experts, and AI-native startups.
Ethically trained AI startup Pleias releases new small reasoning models optimized for RAG with built-in citations - Carl Franzen, Venture Beat
French AI startup Pleias made waves late last year with the launch of its ethically trained Pleias 1.0 family of small language models — among the first and only to date to be built entirely on scraping “open” data, that is, data explicitly labeled as public domain, open source, or unlicensed and not copyrighted. Now the company has announced the release of two open source small-scale reasoning models designed specifically for retrieval-augmented generation (RAG), citation synthesis, and structured multilingual output. The launch includes two core models — Pleias-RAG-350M and Pleias-RAG-1B — each also available in CPU-optimized GGUF format, making a total of four deployment-ready variants. They are all based on Pleias 1.0, and can be used independently or in conjunction with other LLMs that the organization may already or plan to deploy. All appear to be available under a permissive Apache 2.0 open source license, meaning they are eligible for organizations to take, modify and deploy for commercial use cases.
Sunday, May 04, 2025
Exploring model welfare - Anthropic
Human welfare is at the heart of our work at Anthropic: our mission is to make sure that increasingly capable and sophisticated AI systems remain beneficial to humanity. But as we build those AI systems, and as they begin to approximate or surpass many human qualities, another question arises. Should we also be concerned about the potential consciousness and experiences of the models themselves? Should we be concerned about model welfare, too? This is an open question, and one that’s both philosophically and scientifically difficult. But now that models can communicate, relate, plan, problem-solve, and pursue goals—along with very many more characteristics we associate with people—we think it’s time to address it.
Bot Traffic Surpasses Humans Online—Driven by AI and Criminal Innovation - Kevin Townsend, Security Week
AI is helping internet bot herders with greater scale, lower costs, and more sophisticated evasion techniques. Bots on the internet now surpass human activity, with 51% of all internet traffic being automated (bot) traffic. Thirty-seven percent of this is malicious (bad bots), while only 14% are good bots. Much of the current expansion is fueled by criminal use of AI, which is likely to increase. Within the bad bots there has been a noticeable growth in simple, but high volume bot attacks. This again shows the influence of AI, allowing less sophisticated actors to generate new bots, and use AI power to launch them. This follows the common trajectory of criminal use of AI: simple as the actors learn how to use their new capability, followed by more sophisticated use as their AI skills evolve. This shows the likely future of the bot threat: advanced bots being produced at the speed and delivery of simple bots. The bad bot threat will likely increase.
Saturday, May 03, 2025
I Tested 5 AI Assistants—and What I Found Was Surprising - Carmine Gallo, Inc.
Agentic AI Adoption Blueprint Released by Salesforce - Scarlett Evans, AI Business
Salesforce has released what it calls a blueprint for businesses operating in the agentic AI era.The Agentic Maturity Model highlights four key areas for advancement, helping companies move from early deployment of AI tools such as chatbots, to truly autonomous agents that can work collaboratively with humans. The four levels of agentic maturity outlined in the report include chatbots and co-pilots, information retrieval agents, simple orchestration and single domain, complex orchestration and multiple domain, and multi-agent orchestration.
Friday, May 02, 2025
Using AI to predict student success in higher education - Denisa Gándara and Hadis Anahideh, Brookings
As AI becomes more accessible, higher education is increasingly turning to prediction algorithms to inform decisions and target support services. Prediction algorithms can underestimate success for Black and Hispanic students, disproportionately predicting failure erroneously, even when those students ultimately graduate. Bias-mitigation techniques built into model training are more effective than those applied to the data beforehand, but no single method eliminates disparities.
OpenAI says newest AI model can ‘think with images,’ understanding diagrams and sketches - Hayden Field, CNBC
OpenAI released its newest AI model that it said can understand uploaded images like whiteboards, sketches and diagrams, even if they’re low quality. The company called o3 its most advanced model yet and also released a smaller model called o4-mini. OpenAI is racing to stay ahead in generative AI as competitors including Google, Anthropic and Elon Musk’s xAI ramp up development.
Thursday, May 01, 2025
Introducing: The world's fastest Conversational Video Interface for developers - Julia Szatar, Tavus
At Tavus, our mission is to make digital experiences as immersive as human face-to-face interactions by empowering people to leverage their likeness at scale online. Back in March, we launched our breakthrough Digital Replica model, Phoenix, and Video Generation on our developer platform. Today, we’re thrilled to announce: the Conversational Video Interface. Developers can now build rich, realistic, real-time conversational experiences with digital twins on the Tavus platform. Try talking to Carter in our live demo on our homepage.
"The Industry Reacts to o3 and o4!" - Matthew Berman, YouTube
The video discusses the industry's reaction to the release of O3 and O4 AI models:
- O3's High Intelligence: O3 is highlighted for achieving a near-genius level IQ score (136), surpassing competitors like Gemini 2.5 Pro. It shows strong capabilities in iterative tool use and discovering new information [00:09, 01:05].
- O4 Mini's Tool Use: O4 mini demonstrates advanced reasoning by incorporating tool calls (like writing and executing Python code) directly into its problem-solving process [02:58].
- Significant Innovation: O3 is considered a major advancement in AI, comparable in impact to ChatGPT, especially regarding its utility and ability to handle complex tasks [03:37].
- Overall: The release represents a significant step forward in AI, showcasing impressive reasoning, tool use, and problem-solving skills, although some limitations remain [15:03].
(summary provided in part by Gemni 2.5 Pro)