Weekly AI News

Hey it’s Jul,
Greetings and welcome to the fifth edition of “Weekly AI News”!

A new week full of announcements, highlighted by the much-anticipated release of Grok-3 by Elon Musk, claiming it to be the “most intelligent model in the world”—a bold statement that has undoubtedly caught the attention of competitors.

Over at Meta, the goal is clear: to become the top provider of software and sensors for humanoid robots, aiming to create an Android-like ecosystem for robotics. Meanwhile, Google isn’t left behind: its multi-agent “co-scientist” could speed up medical and genetic research, already outperforming some human experts on high-level tests.

As for Microsoft, its new Majorana 1 quantum chip promises unprecedented reliability, with potential AI applications in the medium term. And while Fiverr offers freelancers the chance to train their own AI—riding the wave rather than being overwhelmed by it—the NBA is experimenting with robotic rebounders and virtual coaches. Who would have thought AI would make its way onto the basketball court so quickly?

Happy reading!

🤖⚡ Elon Musk and xAI unveil next-gen Grok-3

Grok-3 is touted as “the smartest AI on Earth,” achieving state-of-the-art results in math, science, and coding. On key benchmarks, both Grok-3 and its mini version outperform Gemini-2 Pro, Claude 3.5 Sonnet, and GPT-4o. xAI’s Colossus supercomputer, equipped with 200,000 H100 GPUs, trained Grok-3, validating the power of large-scale computing. This leap puts two-year-old xAI at the forefront of the AI race, but competition looms with GPT-4.5, GPT-5, and other major players.

⚙️✨ Grok 3 available free for a limited time

xAI has rolled out Grok 3, a significant upgrade to Grok 2, and made it free to try for a short period. Users can explore new features like DeepSearch and Think Mode, while X Premium subscribers get priority access and extras like Voice Mode. Demand is expected to surge, so the free tier might experience server congestion. For now, anyone can test Grok 3’s advanced reasoning capabilities without needing X Premium.

🦾🏗️ Meta invests heavily in AI-powered humanoid robots

Meta aims to be the backbone of the humanoid robotics market, focusing on software, sensors, and AI infrastructure for third-party manufacturers. Ex-Cruise CEO Marc Whitten will lead the new team within Reality Labs. Initially, Meta won’t launch its own branded robot but will integrate its AI and sensor tech into robots for home tasks. The long-term goal is to provide a foundational platform—like Android did for mobile—while managing safety concerns.

🔬🤝 Google’s multi-agent AI co-scientist

Google introduced an AI co-scientist built on Gemini 2.0. It deployed six specialised agents to generate and validate research hypotheses in fields like medicine and genetics. In trials, it uncovered new drug applications and predicted gene transfer mechanisms in days. Early tests show over 80% accuracy on expert benchmarks, surpassing existing AI models and some human experts. Google is rolling out the system via a Trusted Tester Program for researchers worldwide.

🔍💬 You can now ask Gemini about past chats

Gemini can now retrieve your previous conversations without you having to search through them. It’s not truly “unlimited memory” since it searches through past chats rather than retaining all info continuously. ChatGPT may soon introduce a similar feature, indicating a growing trend of AI models offering better recall capabilities.

🎥🗣️ Google rolled out new AI features to Google Meet

Google Meet now includes a scrollable captions history, enabling users to review up to 30 minutes of live and translated captions. This enhancement improves accessibility and collaboration for remote meetings. The update also underscores Google’s continued push to integrate AI tools across its product ecosystem.

⚖️🚫 Anthropic challenges U.S. proposal against Google’s AI funding

Anthropic opposes a government proposal to ban Google from investing in AI startups, arguing it would stifle competition. Google, which has already invested $3B in Anthropic, lost its antitrust case related to online search. Regulators fear tech giants could use AI investments to reinforce their own ecosystems, but Anthropic counters that banning Google would mainly benefit rivals like OpenAI and Meta.

📈🔓 Anthropic launches open-source Economic Index

Based on millions of Claude.ai conversations, the new index measures how AI affects economic activities. Early findings show AI is mostly used for software development and technical writing, with a focus on “augmentation” over full automation. This research offers insight into the real-world impact of AI on labour and productivity.

🔎📝 Perplexity launches freemium Deep Research feature

Perplexity’s new autonomous research agent handles dozens of queries, reads hundreds of sources, and compiles structured reports in under five minutes. It scored 21.1% on Humanity’s Last Exam—beating Gemini Thinking (6.2%) and Grok-2 (3.8%) but trailing OpenAI’s 26.6%. Users can try up to five free Deep Research tasks daily; Pro subscribers get unlimited access. Perplexity CEO Aravind Srinivas took a playful jab at OpenAI’s Sam Altman, claiming his startup “mogged” the competition.

⚛️💻 Microsoft announces new quantum computing chip, “Majorana 1”

Microsoft revealed a breakthrough quantum chip that, thanks to a “topological state” of matter, is more stable than previous technologies. The advancement, published in Nature, could potentially train and run large AI models within years, not decades. While production remains limited, Microsoft aims to eventually offer Majorana 1 on its Azure cloud. Competitors like IBM and Google are also racing to stabilise qubits for real-world quantum applications.

🎮🛠️ Microsoft debuts AI model for video game development

Called Muse, the model generates dynamic game environments that react to player actions, trained on seven years of gameplay footage from Bleeding Edge. Microsoft open-sourced Muse and hosted a prototype on Azure for developers to experiment with. Although currently limited to Bleeding Edge-like visuals, the tech hints at faster development cycles and personalised gaming experiences.

⚗️⏩ Microsoft’s new AI speeds up protein research

BioEmu-1 predicts how proteins fold and move, generating thousands of structures per hour—a 100,000x speedup over traditional methods. Trained on massive molecular datasets, the model maintains lab-level accuracy even for proteins it hasn’t seen. Now freely available to researchers through Azure AI Foundry Labs, BioEmu-1 could dramatically accelerate medical and biotech discoveries.

🌐🔓 Baidu plans to open-source Ernie 4.5 in 2025

Baidu announced it will release its Ernie 4.5 model under an open-source license by June 30, 2025. Ernie 4.0 already supports text, image, and video generation, boasting advanced reasoning. With Ernie 5 slated for late 2025 and DeepSeek-R1 outperforming OpenAI’s O1, the open-source trend continues to gain momentum.

🗺️📢 Mistral’s first region-specific AI

Mistral Saba is a 24B-parameter language model tailored for Middle Eastern and South Asian regions. It supports Arabic and languages like Tamil and Malayalam, enabling culturally relevant content and more natural conversations. Saba offers faster, more cost-effective performance than larger competitors, with both cloud API and local deployment options.

🇫🇷📱 Mistral tops the French App Store

Mistral’s Le Chat app reached 1 million downloads in 14 days, climbing to the top of France’s App Store. In comparison, ChatGPT reached 500,000 downloads in 6 days but initially launched only in the U.S. The milestone underscores the strong public demand for local AI chat solutions in France.

💻💰 OpenAI’s new software engineering benchmark

SWE-Lancer evaluates real-world freelance dev tasks from Upwork—1,400 jobs worth a combined $1 million. It measures both coding quality and project management decisions, simulating “earnings” if a model solves tasks correctly. Claude 3.5 Sonnet took the lead at $400K out of $1M, illustrating the potential for AI to disrupt the freelance economy.

📈🤖 ChatGPT hits 400M weekly active users

OpenAI COO Brad Lightcap confirmed that ChatGPT grew by 33% in under three months, now boasting 400 million weekly users and over 2 million business accounts. The Reasoning Model API usage quintupled since o3-mini’s release. GPT-5, rumoured to be launching soon, will be completely free with no usage caps.

🧬🌳 The largest AI model for biology

Arc Institute and Nvidia unveiled Evo 2, the most prominent AI model for genomic research. It is trained on 9 trillion DNA building blocks spanning 128,000 species and handles sequences up to 1 million nucleotides, predicting cancer-causing mutations with 90% accuracy. With 40B parameters, it’s available via Nvidia’s BioNeMo platform, paving the way for drug design and synthetic biology breakthroughs.

👂🤟 NVIDIA partners with American Society for Deaf Children on “Signs”

“Signs” is an AI-powered platform offering real-time feedback for American Sign Language learners alongside a dataset of 400,000 sign-language video clips. The initiative aims to make ASL more accessible and accelerate learning for both deaf and hearing individuals.

🎨🔄 Pika Labs releases Pika Swaps

Pika Swaps lets users replace any object or character in a scene using text or image prompts—think a genAI version of a Snapchat filter. With over 12 million users, Pika Labs also launched an iOS app enabling creative video editing on the go.

🎙️🌍 Spotify integrates ElevenLab’s AI voice technology

Spotify now allows authors to produce and distribute AI-narrated content in 29 languages through ElevenLab’s voice solution. This could transform audio content localisation and broaden access to global audiences. It also sets the stage for new AI-driven audio experiences.

🗣️⏳ Amazon’s revamped Alexa delayed

Amazon is postponing the launch of its AI-upgraded Alexa—initially planned for late February—due to issues with incorrect answers. A free trial was supposed to go live, offering features like personality customisation and ride-hailing requests. The new timeline suggests it won’t arrive before March 31, with some premium features potentially locked behind a subscription.

💬🤖 Tencent starts integrating DeepSeek’s model into WeChat

Tencent has begun testing DeepSeek’s open-source model in Weixin, the Chinese version of WeChat, supplementing its own Hunyuan AI engine. The move follows DeepSeek’s global rise after its R1 model showcased performance near OpenAI levels at a fraction of the cost. Tencent’s shares jumped on investor hopes that DeepSeek-powered search will bolster WeChat’s AI offerings in China.

🔒📵 South Korea suspends downloads of DeepSeek AI app

South Korea’s Personal Information Protection Commission has halted app-store downloads of DeepSeek over data-collection concerns. The agency says DeepSeek must address local privacy laws before resuming downloads. This highlights the Chinese AI startup's global regulatory challenges amid its rapid user growth.

🚀🧠 Mira Murati’s OpenAI rival: Thinking Machines Lab

Former OpenAI CTO Mira Murati launched Thinking Machines Lab, a mission to create more “understood, customisable, and generally capable” AI through open science. The star-studded team includes experts from DeepMind, Character AI, and Mistral. The lab, which will launch just six months after Murati left OpenAI to explore new directions, promises to regularly publish papers, datasets, and model specs.

💸🤖 Ilya Sutskever’s startup in talks to raise at a $30B valuation

Safe Superintelligence, founded by ex-OpenAI chief scientist Ilya Sutskever, is negotiating a $1B funding round. The deal would boost its valuation from a previous $5B to $30B in a matter of months. Backed by top investors like Andreessen Horowitz and Sequoia, the company remains tight-lipped about its product roadmap, reflecting the intense investor appetite for next-gen AI labs.

🏠🦾 Figure debuts new system for household robots

Figure’s Helix is a Vision-Language-Action model that enables robots to understand voice commands and handle unfamiliar objects. It combines a 7B-parameter “brain” for comprehension with an 80M-parameter control model for precision. Two robots were shown working together to sort groceries using only voice prompts. The rollout signals a step closer to practical humanoid robots in everyday settings.

🤖💸 Nvidia-backed robotics startup Field targets $2B valuation

Field AI, just two years old, develops AI models that empower robots with humanlike actions. It’s reportedly raising hundreds of millions at a $2B valuation, quadrupling last summer’s $500M figure. This momentum underscores heightened investor interest in generative AI for robotics, aiming to boost both speed and accuracy.

💻📌 HP buys AI wearable startup Humane for $116M

After struggles with its AI “pin,” Humane—founded by former Apple executives—has been acquired by HP. The deal will help HP integrate AI features into its printers, PCs, and conferencing tools. Existing pins will cease functioning at month’s end, marking the close of Humane’s initial wearable device ambitions.

🎨🛠️ Fiverr’s AI platform for gig workers

Fiverr Go offers AI tools so freelancers can train personal AI Creation Models on their past work, selling AI-generated versions for an added revenue stream. A $29/month Personal AI Assistant handles routine tasks and client communications. Fiverr is also launching an equity program to reward top performers with company shares, reflecting a push to integrate freelancers into the AI revolution.

📰❌ Over half of LLM-written news summaries have “significant issues”

A BBC study found more than 50% of AI-generated news summaries contain major inaccuracies, such as misstated facts and misquoted sources. Among ChatGPT-4o, Microsoft Copilot Pro, Google Gemini Standard, and Perplexity, Gemini performed the worst at a 60% error rate. The BBC plans further evaluations, stressing the risk of misinformation in AI-based news tools.

🧩📉 Older AI models show signs of “cognitive decline”

Researchers tested ChatGPT, Claude Sonnet, and Gemini on the Montreal Cognitive Assessment. While ChatGPT-4 scored 26/30, older versions like Gemini 1.0 hit only 16, suggesting degradation in complex reasoning over time. The study sparks debate about AI’s reliability in critical fields like medicine if such “cognitive deficits” persist.

🦠⏱️ AI matches decade-long superbug research in days

Google’s Co-Scientist system confirmed a bacterial resistance mechanism that took Imperial College researchers 10 years to uncover. The AI-generated five valid hypotheses, one mirroring the team’s unpublished findings exactly. This milestone underscores how AI can drastically compress the time for scientific breakthroughs, transforming the pace of discovery.

🗞️🤖 The New York Times’s AI for the newsroom

The NYT now allows AI tools for tasks like SEO headlines, brainstorming, and research—but not for full article writing or image generation. Approved software includes GitHub Copilot, Google Vertex AI, NotebookLM, and OpenAI’s API (non-ChatGPT). The paper introduced Echo, an internal summarisation tool, while it continues a copyright legal battle against OpenAI’s data training practices.

🏀🤖 NBA showcases AI and robotics at 2025 All-Star Weekend

The league unveiled various AI-driven robots used by teams like the Golden State Warriors for rebounding, coaching, and player wellness. Examples include A.B.E. for retrieving passes, M.I.M.I.C. for running formations, and K.I.T. for motivational support. These developments highlight how robotics and AI are finding real-world applications in pro sports, primed to become increasingly widespread.

Community

Join AI Whisperer Community!

Ready to take your AI journey to the next level? Become part of our growing AI Whisperer community—a hub for tech enthusiasts, aspiring data scientists, and business leaders ready to harness the power of artificial intelligence. Inside, you’ll find:

• Community: Like-minded individuals keen to grow together

• Curated AI Tools: Discover the latest and greatest tools to boost productivity.

• Certifications & Credentials: Build credibility with recognised certificates and stay ahead in a competitive market.

Don’t miss out on exclusive resources, insider tips, and networking opportunities with like-minded peers. Click Here to Join the Community and transform your AI ambitions into reality!

That's it for this week!
Until next time, stay curious and keep exploring the ever-evolving world of AI!
Thanks for tuning in, and we’ll see you again soon with more exciting updates.

Jul