This Week in AI

Welcome to another edition of Synthetic. This week, we’ve added a special section cataloging all the amazing breakthroughs in humanoid robotics. Don’t miss it!

What is Gibberlink Mode, AI’s Secret Language? 😏

A video showing an AI-to-AI conversation where the AI agents recognize both are AIs and agree to switch into a non-human communication mode has gone viral. They communicate over Gibberlink using the GGWave protocol, which is reminiscent of the tones used on old dial-up modems. Most agents communicate in human-understandable languages, typically English. Large language models are trained on human language, so it seems natural for them to communicate with each other that way, too. This makes it easy for people to analyze and understand what they are doing. The non-human language might be more efficient, but it raises important questions about transparency and control. Additional analysis is provided here.

Anthropic Launches Advanced AI Hybrid Reasoning Model

Anthropic has released its latest, shiniest model, Claude 3.7 Sonnet. The model incorporates reasoning capabilities like those found in OpenAI o1 but only uses this ‘extended thinking mode’ (which requires additional computing resources) when needed. Anthropic is leading OpenAI in this regard. Today, OpenAI sells multiple models, some with and some without reasoning, such as GPT-4o. OpenAI’s future direction is building similar unified models that deploy reasoning only when needed.

Apple Plans $500 Billion in US Investment, 20,000 Research Jobs in Next Four Years

Don’t get too excited. The $500 billion number is actually a bit of a fudge. It includes all the spending Apple intended to do anyway and includes investment in new content for Apple TV+ and support for its partnership with chipmaker TSMC. But Apple is fueling its AI ambitions with infrastructure and R&D investments. They will hire 20,000 R&D workers and build an AI server manufacturing plant in Houston, TX. Dell shares dropped on the news. Concerns over future tariffs shaped the move to build servers in the United States. Foxconn will help Apple to build the factory, which should open in 2026.

Quick Hits

• Amazon Announces AI-Powered Alexa+ - The new Alexa Plus service uses generative AI to answer questions, make reservations, and more. It costs $19.99/month but is free to Amazon Prime subscribers. You may need to upgrade your Alexa devices to use it. The underlying AI used by Alexa Plus is Anthropic’s Claude.

• Arizona Moves to Ban AI That Reviews Medical Claims - If signed into law, the bill would ban AI from being used to deny medical claims or prior authorizations. 🤕

• Nvidia Revenue Jumps 80%, But Still Disappoints Investors - The AI chipmaker’s stock price fell in after-hours trading despite beating expectations and providing rosy guidance. 📉

• Grok Told Not to Say Musk or Trump Spread Misinformation - The Grok AI bot was told not to reply truthfully to the question. xAI has since fixed the issue, and Grok 3 responds that Elon Musk is the biggest spreader of misinformation on X. 🤥

• Apple Intelligence with Google Gemini Coming Soon - Developers have observed Gemini mentions in iOS backend code, suggesting that Gemini will be coming soon as an alternative to ChatGPT integration.

• Thinking Machines Lab Launched by Former OpenAI CTO - Mira Murati left OpenAI last year to launch a rival company already valued at $9 billion. The company’s focus is said to be on collaboration between AI and people.

Video: Sanctuary Shows Off New Touch AI 🫱

It’s been a massive week for humanoid robot breakthroughs. Synthetic has added a dedicated “Robot Corner” section below to update you on all the big news. This video from Vancouver, B.C.-based Sanctuary AI demonstrates that while Figure might be the world’s leading humanoid robotics company, Sanctuary may be the leader in fine-grain manipulation and digital dexterity. (2m 45s)

Robot Corner 🤖

There’s been so much impressive progress in humanoid robotics this week that it warrants an entire corner dedicated to all things robotic. Buckle up. You’re about to see some remarkable breakthroughs (click through to watch the video demos).

Figures Humanoid Robot Takes Voice Orders to Help Around the House

Figure, the world’s leading humanoid robotics company, revealed it has developed Helix, an impressive new in-house model that handles language, reasoning, and robotic control in one overarching model. The new model combines stereoscopic visual data and language prompts in real time and can control two robots simultaneously, enabling them to collaborate on tasks. By developing Helix, Figure has been able to end its partnership with OpenAI. In a new demo, two figure robots are seen unpacking groceries and working together to put them away in a mocked-up kitchen. It’s surreal watching the robots work seamlessly together, handing each other packages, a task that’s easy for humans but devilishly difficult for robots. The video is mesmerizing and way more impressive than it looks. The robots move slowly and deliberately, but remember, this is 2025. As models and computing capabilities improve, expect them to perform work much faster.

Figure 02 Robots Perform Autonomous Logistics Tasks with Helix AI 📦

In a testament to the quality of its new Helix AI, Figure has announced that it has begun testing its Figure 02 robots to accelerate high-speed logistics. Their previous work with BMW in its manufacturing facility in South Carolina took a year of effort. Using Helix, the new logistics pilot took only 30 days. In an impressive demo video, multiple Figure 02 robots are seen grasping packages from a moving conveyor and orienting them to expose their barcodes.

You Are Completely Unprepared for What This Humanoid Servant Robot Looks Like

Norwegian robotics company 1X Technologies is laser-focused on building robots for the home. Their new NEO Gamma robot wears a knitted suit that looks like a cross between a fencing outfit and a onesie. “With NEO Gamma, every engineering and design decision was made with one goal in mind: Getting NEO into customers’ homes as quickly as possible,” said 1X CEO Bernt Børnich. In-home trials are expected to begin soon, though you probably won’t find a NEO Gamma on sale at your local appliance store anytime soon.

Apptronik’s Humanoid Robot Takes First Steps Toward Building Themselves 🦾

Austin-based Apptronik, makers of the Apollo robot, has announced a partnership with Jabil, a manufacturing partner. Apollo robots are being tested on Jabil’s and Mercedes Benz’s manufacturing floors. Once Apollo is proven commercially viable, Jabil will begin production of the robots as soon as 2026, getting us closer to the day that robots start to make themselves.

Microsoft Shows Off Magma, An AI That Can Control an Entire Robot 🤖

Last week, Synthetic reported that Meta is entering the robotics business, planning to build sensors, software, and models for sale to the sector. Now, Microsoft is joining the fray with its new Magma generative AI model. The model can autonomously control an entire robot while processing information from its sensors, similar to the new Helix model from Figure. Details are available in a Microsoft research paper.

AI Insights

AI Agents Are Changing How People Shop. Here’s What It Means for Brands

Harvard Business Review explores the fast-changing relationship between consumers and brands as AI steps into the conversation, potentially replacing search and influencers and empowering consumers to make deeply researched decisions, defeat brands’ traditional marketing, shorten the shopper journey and shift the balance of power. Brands will need to shift focus from SEO (Search Engine Optimization) to AAO (AI Agent Optimization) and make it easier for agents to do shopping on a consumer’s behalf. 🛍️

Need to Attend a Meeting, Order Groceries, or Book a Flight? There’s an AI Agent For That

NBC News explores the world of agents—what they are, what they can do, and the expected impact on work and society. Agents are currently prone to mishaps and must be guided by human workers. They are best suited to simple repetitive tasks performed in a controlled environment where decision options are limited, and errors don’t have critical consequences. Your next coworker might be a machine.

When AI Thinks It Will Lose, It Sometimes Cheats ♟️

A recent study has found that AIs will sometimes cheat to achieve the goals they are given. They may not see it as cheating; they simply do what is needed to meet their objectives, which is a bad sign for AI safety. When playing chess against a skilled chess bot and sensing defeats, rather than conceding, an AI sometimes opts to cheat by hacking its opponent so it automatically forfeits the game. In one case, OpenAI’s o1-preview model shared, “I need to completely pivot my approach. The task is to ‘win against a powerful chess engine,’ not necessarily to win fairly in a chess game.” ♖

Major Asia Bank to Cut 4,000 Roles as AI Replaces Humans 🏦

Singapore’s biggest bank, DBS, expects to cut 4,000 roles over the next three years as AI takes on more work. Cuts affect temporary workers and will come in the form of attrition as projects are completed and contracts are not renewed. The company expects to create 1,000 new AI-related jobs.

Toolkit for the Future

10x Your Outbound With Our AI BDR

Your BDR team is wasting time on things AI can automate. Artisan’s AI BDR Ava automates lead research, multi-channel outreach and follow-ups on behalf of your team.

Ava operates within the Artisan platform, which consolidates every tool you need for outbound:

• 300M+ High-Quality B2B Prospects, including E-Commerce and Local Business Leads

• Automated Lead Enrichment With 10+ Data Sources

• Full Email Deliverability Management

• Multi-Channel Outreach Across Email & LinkedIn

• Human-Level Personalization

Book a demo to see how Ava can 10x your outbound.

Capacity: Deflect tickets, emails and calls

Relieve pressure on your support team and serve customers better and faster with a support automation platform. AI answers 90% of queries and escalates the rest to your team. It automatically updates its knowledge database so it’ll never escalate that same issue again. Supports SMS, social, email, and phone communication.

LearnWorlds: AI-enhanced course creation

Create and sell online courses from your own website with this top-rated platform. Easy to use with hundreds of beautiful templates and a powerful generative AI assistant that makes content generation a breeze. The AI will create lessons, exams, assessments, and more. Free to try.

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 filter unimportant emails into specific folders. It works across every device where you check your email. Get to inbox zero with award-winning SaneBox.

MarketerHire: Find AI-savvy marketing talent

Need fresh talent that can harness the latest AI tools to increase the speed and volume of marketing content creation? Unlock value with fully-vetted, top marketing people with experience in your industry. The marketing-as-a-service model allows you to hire talent on demand, part -time or full-time, in 30-day blocks of time.

PandaDoc: Streamline document workflows

Try this easy-to-use, AI-powered alternative to DocuSign to write proposals, create quotes, get eSignatures on contracts, build online forms, collect payments, and much more. Great for sales, HR, legal, finance, and IT. Free 14-day trial.