This Week in AI

OpenAI Plans to Charge $20,000 a Month for PhD-Level Agents

OpenAI has told investors it plans to create massive new revenue streams by launching new AI agents. These agents are designed to support the efforts of knowledge workers. A ‘low-end’ agent will cost $2,000 per month and will be aimed at high-income knowledge workers. Mid-tier agents designed for software development might cost $10,000 a month, and high-end agents who work as Ph.D.-level researchers could cost $20,000 per month. Longer term, OpenAI expects about a quarter of its revenue to come from selling agent products. OpenAI investor SoftBank has already committed to spend $3 billion on agents from OpenAI this year. Today, OpenAI’s top level of capability, ChatGPT Pro, costs $200 a month but uses so much computing power to operate that OpenAI loses money on the deal.

Synthetic’s Take: The world of work is about to change, and your new coworker could be a machine. Digital workers will toil for electrons rather than dollars and cents, and agentic AI will flood the workplace in 2025 and 2026. Companies like OpenAI plan to cash in on agents and capture a significant fraction of every company’s labor budget. The blended workforce of the near future will consist of people, agents, and robots working side by side.

OpenAI Launches GPT-4.5 to Thunderous Lack of Enthusiasm

OpenAI seems to enjoy launching new technologies on Thursdays, moments after Synthetic’s weekly missive hits your inbox. Last week was no exception as OpenAI finally revealed its new monster model, GPT-4.5. For months, OpenAI has signaled that its latest model, previously codenamed Orion, was not its best work and was disappointing compared to hopes and expectations. GPT 4.5 has received mixed reviews, with some calling it a lemon and highlighting its poor coding performance. The model is much larger and more GPU and energy-hungry than the previous GPT-4 generation, making it expensive. GPT-4.5 has enhanced writing capabilities, higher EQ, deeper world knowledge, and better intuition and creativity. But it still hallucinates (37% of the time) and has no reasoning layer like its brothers o1 and o3-mini. It’s being seen by many as a placeholder before OpenAI releases GPT-5, which will integrate reasoning and is expected to take the crown as the smartest model yet available when it launches in the next couple of months.

Synthetic’s Take: Once GPT-5 launches, expect GPT-4.5 to be quietly retired. That should tell you all you need to know about whether you should start using it.

Figure Will Start Testing Robots in the Home This Year

Figure has made such rapid progress with its robot technology that it will accelerate its plan to launch robots into homes. Recent breakthroughs with its new Vision-Language-Action model, Helix, essentially an operating system for robots, have given Figure the confidence that its Figure 02 robot is ready for alpha testing in homes later this year. Other robotics companies targeting the home include 1X, Apptronik, and Tesla. Apple is rumored to be exploring the opportunity and actively doing R&D. What will you have your home robot do for you? Wash the dishes, do the laundry, mow the lawn, cook dinner, walk the dog, grab you a beer, shake you a martini? 🍸

Quick Hits

• Apple Delays Siri’s AI Upgrade to 2027! 🐌 - Apple Intelligence looks more like Apple Dumb as Apple exposes its chronic underinvestment in AI and signals a full-on AI emergency for the company. While Amazon is set to launch its AI-powered Alexa Plus service this month, the AI reboot of Siri will likely be pushed into 2027.
  Synthetic’s Take: It’ll probably be great when it comes, but by then, artificial general intelligence might be here, perhaps making the new Siri irrelevant. Bottom line: Apple is woefully behind.

• Deutsche Telecom Will Launch New AI Phone - The T-Mobile parent company’s new phone will feature Perplexity AI Assistant on the Lock Screen, launch in the second half of 2025, and roll out across Europe in 2026.📱
  Synthetic’s Take: This is the first sign of our prediction that the app-first smartphone model will give way to an AI-first model and shift the primary interface.

• Microsoft Launches Dragon CoPilot AI Assistant for Healthcare - The assistant takes clinical notes and supports multiple languages. Its technology came from Nuance, a voice dictation company Microsoft bought in 2021. In trials, clinicians reported less burnout when using Dragon, and 93% of patients reported a “better overall experience.” 🧑‍⚕️

• Amazon Gets into Reasoning Model Business - Rumors are flying that Amazon plans to compete head-to-head with OpenAI, Google, and Anthropic and will launch its own reasoning model, branded ‘Nova.’

• AI Chip Leaders Broadcom, AMD, and Nvidia Rumored To Try Intel Manufacturing - The chip giants have no manufacturing capability and traditionally rely on TSMC to make their chips. Intel’s new 18A process is seen as competitive, with the 14A process (meaning transistors are just 14 atoms wide!) not far behind it. AI chips will keep getting more capable as Moore’s Law roars onward.

• OpenAI CEO Says They Are Out of GPUs - As OpenAI builds its newest models, it’s running out of computing capacity, says CEO Sam Altman. This explains the ridiculously high price to access their latest model, GPT-4.5, which is huge and expensive to run. Access to GPT-4.5 costs 30 times the price of using GPT-4o. OpenAI is using pricing to control demand as it doesn’t have the computing power to run it at scale.
  Synthetic’s Take: Nvidia is selling every chip it can make. Demand will outstrip supply for the foreseeable future despite the mistaken view that models like DeepSeek R1 signal a reduced need for powerful computing. Altman’s statement reinforces that even the biggest names in AI can’t get enough of what Nvidia has to offer.

Videos: Humanoid Robots, Trillions of Dollars, and the Future of Humanity

Explore the frontier of humanoid robotics in this wide-ranging interview with serial entrepreneur and CEO of Figure, Brett Adcock. Adcock’s company is only three years old, yet it’s at the forefront of humanoid robotics today. Adcock predicts a gigantic future market for robots in commerce, the home, and even off-world. If you’re interested in the future potential of humanoid robots, this is well worth your time. (57m 46s)

AI Tech and Innovation

World’s Largest Call Center Uses AI to ‘Neutralize’ Indian Accents

Teleperformance SE, the planet’s largest call center operator, uses AI from startup company Sanas to improve audio quality and mask the accents of people delivering support over the phone. The technology is designed to make it easier to hear and understand people with thick Indian or Filipino accents and works with zero latency. The AI also masks background call center sounds. It plans to deploy the technology in Latin America next. The online demo seems impressive.

AI Models Prone to Overthinking, Boosting Costs

The key to building smarter AIs is to have them think less, not more. That’s the conclusion of a research paper by the University of California Berkley, ETH Zurich, Carnegie Mellon University, and the University of Illinois. Large language models that reason (so-called Large Reasoning Models [LRMs] like OpenAI’s o1, DeepSeek R1, or Alibaba’s QwQ-32B) are prone to getting stuck in their own heads and overthinking problems. This wastes computing time, energy, and money. The paper’s authors made their findings available in hopes of helping AI researchers address the issue.

AI Insights

AI Movie Studio Makes Near-Studio-Quality Movies for Under $500K

New AI-powered movie studio, Staircase Studios, has released the first five minutes of its first feature-length picture, The Woman With Red Hair. The studio was founded by a producer of the Divergent movie series and its first film boasts moviemaking talent from Pixar and Marvel. Moviemaking is set to transform as AI video generation improves over the coming years. You can preview the movie here.

OpenAI Researcher Says Soft Skills Aren’t Going Anywhere

Karina Nguyen, a researcher at OpenAI, said creativity and emotional intelligence remain some of the hardest things to teach AI. Nguyen also discussed the goal of AI research labs, which is to automate AI research itself. “It’s kind of scary, I’d say, which makes me think that people management will stay. It’s one of the hardest things to do—emotional intelligence, with the models, creativity itself is one of the hardest things.”

Synthetic’s Take: To remain robot-proof in the workplace for as long as possible, double down on your humanity and focus on developing the skills that AIs lack.

Toolkit for the Future

HireLogic: Turbo-charge job interviews with AI

Invite the HireLogic AI to your next phone, video, or in-person interview. It analyzes the conversation, provides specific insights, and highlights the interviewee’s strengths, skills, and any potential concerns. It creates a full interview transcript so you can focus on the conversation. Free to try.

Notion: Organize and share team information

Notion helps you focus on the things that actually matter. Organize information, thoughts, to-do lists, calendars, articles, tasks, notes, and team discussions in one well-organized space. Use Notion AI to create content, translate languages, and ask your notes questions.

SaneBox: Get rid of inbox clutter

Focus on the email that matters. Deep Clean automatically identifies old, unimportant emails so you can delete them. Snooze emails until a better time and easily track emails that need a response. Automatically filters unimportant emails into specific folders. It works across every device where you check your email. Get to inbox zero with the award-winning SaneBox.

ElevenLabs: Generate voices, music & sounds

Recording, editing, and refining audio demands more time than most entrepreneurs can spare. Generate voices, high-quality music, and sound effects efficiently and cost-effectively with a single click. Create multilingual voice tracks in up to 29 languages to reach new markets. Try it for free.

Seamless.AI: Boost your sales

Find phone numbers, emails, and company org charts to reach the people you need to. Fill information gaps in your contacts, auto-generate follow-up emails, and identify new target markets. Get 50 free credits and discover why over 500,000 companies use Seamless to grow their business.