Good morning. It’s Wednesday, March 5th.

On this day in tech history: In 1924, The Computing-Tabulating-Recording Corporation (CTR) officially rebranded as International Business Machines Corporation (IBM).

In today’s email:

• Google’s Gemini w/ Vision Coming Soon

• The Most Realistic AI Voice Yet

• Altman Hints At Image Gen Upgrade

• 6 New AI Tools

• Latest AI Research Papers

You read. We listen. Let us know what you think by replying to this email.

Google to launch Gemini with Vision in March for AI-powered live video analysis

Google is rolling out live video analysis and screen-sharing features for Gemini as part of the Google One AI Premium Plan. The update allows users to stream video from their smartphone cameras or share their screens for real-time AI-powered insights. Initially exclusive to Android devices, the features support multiple languages and enhance Gemini’s ability to interpret visual content.

This expansion aligns with Google’s broader vision for multimodal AI, leading up to "Project Astra," an assistant designed to process text, video, and audio in real-time. While Astra’s full rollout remains uncertain, these incremental updates suggest Google is steadily embedding multimodal AI into everyday interactions.

Google has also added lockscreen widgets to its Gemini AI assistant on iOS and iPadOS, enabling instant access to key features like text prompts, live conversations, voice commands, and image analysis. With a full Siri overhaul still years away, Google is capitalizing on the gap. Read more.

Related Story: Google upgrades Colab with an AI agent tool

Sesame’s AI Voice Demo Stuns With Realism

Sesame AI’s Conversational Speech Model (CSM) delivers strikingly human-like voices, mimicking breath sounds, chuckles, and self-corrections. Built on Meta’s Llama architecture, it processes text and audio in a single-stage transformer model, enhancing realism beyond traditional text-to-speech.

The demo, featuring voices “Miles” and “Maya,” has impressed users while raising concerns over emotional attachment and deepfake risks. Blind tests show CSM’s speech rivals human recordings, though real voices still hold an edge in context. Its ability to roleplay dynamic personalities, including aggressive tones, sets it apart from competitors.

Sesame plans to expand language support, scale its models, and open-source key components. Read more.

Altman Hints at Major Image Generation Upgrade

OpenAI CEO Sam Altman announced that GPT-4.5 will roll out gradually to Plus-tier users over several days. He stated that an immediate full release would have required stricter rate limits, and the team expects high usage.

In a separate response, Altman hinted at significant improvements to ChatGPT’s image generation. When a user complained about declining quality, he replied that they would soon be "wild with joy," suggesting an upcoming upgrade. OpenAI has not provided a timeline for these enhancements. Read more.

• Stability AI and Arm bring offline AI audio generation to smartphones

• Amazon expands AI across its ecosystem, strengthens AGI and compute strategy

• OpenAI’s NextGenAI funds AI research at Harvard, MIT, and Oxford

• Cohere claims its new Aya Vision AI model is best-in-class

• Cortical Labs unveils CL1, the first biological-silicon hybrid computer

• Contextual AI’s GLM beats GPT-4o with unmatched accuracy

• Claude 3.7 Sonnet joins ElevenLabs AI for advanced reasoning and voice AI

• Microsoft's Dragon Copilot AI listens in on doctor-patient talks to automate medical paperwork

• Salesforce launches an AI agent skills marketplace with 200+ initial partners

• T-Mobile’s parent company and Perplexity team up for AI-first smartphone under $1K

• Nvidia CEO Jensen Huang says its US AI chips are around "60 times" faster than Chinese counterparts

• Federal judge denies Elon Musk’s request to block OpenAI for-profit conversion—but she'd be happy to hold a trial

• ASLP Labs debuted DiffRhythm, an open-weights model capable of generating 4-minute songs with vocals

• Nvidia-backed cloud firm CoreWeave to acquire AI developer platform Weights & Biases

• Anthropic raises $3.5 billion in new funding, valuing the AI company at over $60 billion

• People are using Super Mario to benchmark AI now

• LlamaIndex launches a cloud service for building unstructured data agents

• Meta researchers generate photorealistic avatars from just four selfies

• US' light-powered AI camera processes images 99.4% faster without electricity

• China: Interactive humanoid robots take over police patrol duties in streets

• A game developed via “Vibe-coding” brings $52,000/Month

📄 Visual-RFT: Visual Reinforcement Fine-Tuning

📄 Difix3D+: Improving 3D Reconstructions with Single-Step Diffusion Models ​​

📄 Phi-4-Mini Technical Report: Compact yet Powerful Multimodal Language Models via Mixture-of-LoRAs

📄 MPO: Boosting LLM Agents with Meta Plan Optimization 

📄 Mask-DPO: Generalizable Fine-grained Factuality Alignment of LLMs

Thank you for reading today’s edition.

Your feedback is valuable. Respond to this email and tell us how you think we could add more value to this newsletter.

Interested in reaching smart readers like you? To become an AI Breakfast sponsor, reply to this email or DM us on 𝕏!