- Synthetic
- Posts
- GPT-4.5, GPT-5, and AGI are coming š§
GPT-4.5, GPT-5, and AGI are coming š§
Plus, When Capital No Longer Needs Labor, How Does Labor Gain Capital? š°
Subscribe to Synthetic
The weekās most interesting and relevant AI news and analysis
This Week in AI

In the first few phases of the ongoing industrial revolution, we built mechanical muscles. Now, we are building mechanical minds.
In his latest blog post, OpenAI CEO Sam Altman said that AGI is ācoming into viewā and that this makes him enthusiastic and worried about the future. Specifically, he imagines a world where we cure all diseases, and each realize our creative potential. He says, āIn a decade, perhaps everyone on earth will be capable of accomplishing more than the most impactful person today.ā He made three observations:
The intelligence of an AI model roughly equals the log of the resources used to train it. (e.g., scaling laws still hold, though you have to spend more and more to get linear gains)
The cost to use a given level of AI falls about 10x every 12 months, and lower prices lead to much more use (e.g., Jevons Paradox in action)
The socioeconomic value of linearly increasing intelligence is super-exponential.
He admits that our economy and society are about to undergo huge change and that AGIās benefits may not be widely distributed. He calls for proactive policies and technological advancements that democratize access to AIās power and make plans to integrate it into society.
In another post this week, Altman says OpenAI will release its GPT-4.5 and GPT-5 models within āweeksā and āmonths.ā GPT-4.5 was previously known internally by the codename āOrion.ā GPT-5 will integrate many other OpenAI technologies, including the reasoning model, o3, which it plans to cancel as a separate release.
Syntheticās Take: OpenAI has decided that reasoning capabilities are the future of all its models. Reasoning improves capabilities and accuracy (see the hallucination chart below), so they will integrate it across all their offerings and simplify their product line.
At the Paris AI Summit this week, co-hosted by France and India, France stepped up and announced a $112B AI investment package, characterized by some as their answer to the U.S.ā Stargate program. The French government also pledged to build a nuclear power plant to power its AI ambitions. U.S. Vice President, representing the U.S. government, railed against āexcessiveā AI regulation in a rebuke to Europe and told summit attendees that the U.S. would dominate in AI. By contrast, the Chinese Vice Premier offered to share AI achievements, work with other countries to safeguard security, and build a ācommunity with a shared future of mankind.ā Anthropic CEO Dario Amodei released a statement imploring democratic societies to accelerate AI development to prevent authoritarian countries from using it for global dominance. In the summitās final communique, 60 countries (including France, India, China, and Canada) agreed to a series of voluntary commitments to make emerging AI technology inclusive and sustainable. The UK and U.S. refused to sign the declaration. This is a complete list of the summitās announcements.
Quick Hits
An Elon Musk-led Consortium Made a Hostile Bid to Buy OpenAI for Cash - Sam Altman told them it was ānot for saleā and where to sling it. āļø
Viral AI-Generated Video Shows Celebrities Protesting Kanye West - None of the celebrities permitted their likenesses to be used, leading to Scarlett Johansson demanding limits on Deepfakes.
US Navy Launches First AI-Powered Warship - The ship uses AI to monitor its hull, mechanical and electrical systems. It scans 10,000 sensor readings per second to generate maintenance recommendations. āļø
Google Machine Learning Tech Will Estimate a Userās Age - Google says the technology will help it provide age-appropriate experiences.
Adobe Launches New Video Tool to Compete with OpenAI and Google - their new Firefly Video Model competes with OpenAIās Sora and Googleās incredible Veo 2. š„
YouTube Will Let You Put AI-Generated Video Clips in Shorts - It will use Googleās new Veo 2 model as part of their Dream Screen tool. šŗ
Google DeepMind CEO Demis Hassabis Says DeepSeek Is Chinaās āBest Workā So Far - but it shows āno actual new scientific advance.ā
AllState Says Most Communications About Insurance Claims Are Done With AI - Claim agents review emails but donāt write them. AllState PR people are trying to backpedal on the news.
New Law Would Make It Illegal to Use DeepSeek - punishable by 20 years of prison time. Wow, lawmakers really donāt want you to use it! āļø
DeepSeek Likes to Hallucinate - In a recent study, DeepSeekās model hallucinated over 14% of the time, compared to other leading models with hallucination rates just a fraction of that. šµāš«

Video: Google DeepMind CEO Wants To Build A Virtual Cell to Push Boundaries of Medicine
The always-impressive Demis Hassabis talks to Alex Kantrowitz about his ambition to build on his Nobel-prize-winning work with AlphaFold to build a virtual cell using AI as an underlying platform. Within five years, we may be able to use virtual cells to help us understand disease paths and develop new drugs and cures, understand aging, and improve human health. DeepMind is also exploring how AI can help us understand genetic mutations and predict their impact on the body. AI may help us to understand the aging process, interrupt it, and develop rejuvenation therapies. Watch this video if you want a dose of the future delivered by a pragmatic optimist with his finger on the pulse of the future of AI and medicine. (8m 40s) š§¬š¬š§«
AI Tech and Innovation

Get ready to live in a world of AI agents
AI darling Snowflake has launched new Cortex AI agents as part of a service to simplify data integration, retrieval, and processing. These agents use Anthropicās Claude and are available for public preview now. They are built on top of Snowflakeās Cortex Analyst and Cortex Search services and can plan tasks, explore options, and execute them using tools. The agents use reflection after they use each tool to evaluate results and determine next steps. š¤
Syntheticās Take: 2025 will be the year of agents, and you should expect to see many announcements like this over the coming weeks and months. Everybody is about to get their agent on!!
Frontier AI lab Anthropic has released an Economic Index that analyzes millions of conversations with its Claude AI to draw conclusions about how people use it in the workplace. Its initial report includes the following insights:
AI is most commonly used in medium to high-wage roles.
The majority of current AI usage (57%) is focused on augmentation and collaboration. The remaining 43% is focused on task or workflow automation.
About one-third of users use AI in at least 25% of their tasks, but only 4% use it across 75% of their tasks.
Specifically, itās considering how its models should handle controversial topics, user customization, and āintellectual freedom,ā which is OpenAIās term for usersā ability to explore and debate ideas without arbitrary restrictions. OpenAI has expanded and updated its model spec, which defines how its models should behave, and it is making it free for anyone to use or modify.
AI Insights

How will we live good lives when machines do all the work? The article below explores the issue and suggests some solutions.
Emad Mostaque is the former CEO of Stability AI and a controversial character in the AI world. His thoughtful post explores the likely impact of artificial general intelligence on work, the economy, and society. It emphasizes the need for policies that democratize access to AI technologies, ensure equitable distribution of benefits, and address job displacement challenges. He expresses the need to develop new economic models prioritizing human well-being and environmental sustainability in a post-labor economy, including universal basic income and/or public ownership of AI infrastructure. Mustaque proposes five strategies for economic adaptation:
Control of Key Resources: Data pipelines, proprietary algorithms, and computational infrastructure constitute critical bottlenecks in the AI economy. Policymakers and entrepreneurs must prioritize democratizing access to these essential resources.
Data Ownership: Individuals should assert ownership over their personal data and earn royalties or dividends when such data is monetized.
Universal Basic Capital: Governments could distribute equity in AI-driven enterprises to ensure that automationās dividends benefit society collectively rather than exclusively enriching a privileged few.
Universal Basic AI: Inspired by the Intelligent Internet framework, this approach advocates for equitable access to AI tools for education, healthcare, and professional development. Subsidizing AI systems could mitigate socioeconomic divides.
Entrepreneurial Opportunities: By drastically reducing operational costs, AI empowers individuals to launch ventures with capabilities previously reserved for large organizations.
The big labs are all releasing their first batch of agents, AIs that can handle more complex tasks and execute those tasks using your computer's browser. These models are experimental but point to whatās next in computing. These early models can handle simple tasks such as online shopping and booking flights, but they come with significant limitations and some safety concerns.
Former New York Times columnist and Nobel-winning economist Paul Krugman writes that the AI boom is much like the dot-com bubble but may not end in a massive crash. He says that the end game may differ since AI is dominated by major players already well-placed to exploit network effects and operate as quasi-monopolies. He suspects the main risk is that the magnificent seven over-invest and fail to generate adequate returns.
Toolkit for the Future

Generate ad banners, text, images, and video that outperform those of your competitors with better conversion and click-through rates. Get AI-powered creative insights and analyze your competitorsā campaigns to stay ahead of the curve. Free to try.
Rapidly build and deploy specialized AI workers with an intelligence infrastructure that gives everyoneāfrom developers to non-technical teamsāthe power to build with AI. MindStudio already powers over 100,000 AI workers at companies including Adobe, Google, IBM, and Meta. The easy-to-use, no-code, graphic workflow makes it easy for anyone to build an AI coworker. A free level is available.

This all-in-one AI-powered tool helps you stay on top of your spending and investments and offers personalized guidance and insights on managing your credit and personal finances. Thereās a conversational chat interface, and new features are added each month.
Automate Prospecting Local Businesses With Our AI BDR
Struggling to identify local prospects? Our AI BDR Ava taps into a database of 200M+ local Google businesses and does fully autonomous outreachāso you can focus on closing deals, not chasing leads.
Ava operates within the Artisan platform, which consolidates every tool you need for outbound:
300M+ High-Quality B2B Prospects
Automated Lead Enrichment With 10+ Data Sources Included
Full Email Deliverability Management
Personalization Waterfall using LinkedIn, Twitter, Web Scraping & More
Turn customer feedback into action with powerful AI-powered surveying and analytics tools. These tools allow you to measure experience across the customer journey, capture key metrics, convert feedback into growth opportunities, enhance customer satisfaction, and turn customers into loyal advocates. Trusted by brands including American Express, SAP, and Accor, and free to try.
Help your sales team stay focused on customers during meetings. Laxis captures attendeesā comments verbatim and flags items for follow-up. It auto-generates meeting summaries and follow-up emails in seconds, quickly identifying customer requirements, pain points, and action items. Free to try.
Extract data from websites and place it in a spreadsheet that fills itself and notifies you of any changes. Pre-built robots make it a breeze to get started. Extract whatever you need: job listings, property details, TikTok videos, hotel info, or competitor pricing. It's free to try, and no coding is required.