MIT Technology Review Insights has published a playbook for crafting an AI strategy

Highlights

🔹 AI's Economic Impact: AI is projected to significantly boost global GDP by 14% by 2030, contributing an estimated $15.7 trillion.

🔹 Automation of Mundane Tasks: Research by the Oxford University indicates that 40% of routine tasks could be automated by AI by 2030.

🔹 Substantial AI Investment: Investment in AI is expected to reach $200 billion by 2025, according to Goldman Sachs.

🔹 Widespread AI Influence: Experts assert that AI will impact every job and function within organizations. “No job, no function will remain untouched by AI,” says SP Singh, senior vice president and global head, enterprise application integration and services, at technology company Infosys.

🔹 Challenges in Scaling AI: Despite optimistic predictions, only 5.4% of U.S. businesses had fully integrated AI into their operations by 2024.

🔹 Barriers to Enterprise-Wide AI Adoption: Moving from AI pilots to full deployment requires strategic changes in #infrastructure, #datagovernance, and supplier ecosystems.

🔹 Importance of Strategic Planning: Organizations must address uncertainties in AI performance and ROI to scale AI across business functions effectively.

🔹 Rising AI Readiness Spending: Companies are planning to significantly increase spending on AI-related activities, including #data readiness and platform modernization.

🔹 Data Liquidity as a Key Factor: The ability to access, combine, and analyze data seamlessly is critical for effective AI deployment.

🔹 Governance and Security Concerns: #Governance, #security, and #privacy issues are major obstacles, slowing down AI deployment for 45% of companies.

🔹 Data Quality Issues: Half of the respondents consider data quality as a significant barrier to AI deployment, especially in large firms with complex IT infrastructure.

🔹 Cautious AI Adoption: Nearly all organizations (98%) prefer to delay AI deployment to ensure it is implemented safely and securely, with larger companies particularly concerned about governance and security.

Cerebras Co-Founder Deconstructs NVIDIA Blackwell Delays

From intricate interposer designs to alignment issues and thermal expansion complications, Cerebras Co-Founder and Chief System Architect Jean-Philippe Fricker provides a detailed look into the hurdles faced by GPU architectures as they try to go bigger.

❗️This work marks the beginning of a new era of automated, open-ended scientific discovery.

Sakana AI introduced The AI Scientist: The world’s first AI system for automating scientific research and open-ended discovery!

From ideation, writing code, running experiments and summarizing results, to writing entire papers and conducting peer-review, The AI Scientist opens a new era of AI-driven scientific research and accelerated discovery.

Paper.
Code.

Google launched Gemini Live, a mobile conversational AI with advanced voice capabilities and 10 different voices.

It's available to all Gemini Advanced users on Android, but multimodal capabilities are still coming later this year.

Google also launched its Pixel Buds Pro 2 globally, with a custom Tensor A1 chip to power Gemini functionality.

This puts the Gemini Live voice assistant directly in your ear for a "hands-free, eyes-free virtual AI assistant" anytime you need it

xAI launched new model Grok-2

Researchers demonstrated a speech neuroprosthesis that decodes the attempted speech of a man with ALS into text with 97.5% accuracy, enabling him to communicate with his family, friends, and colleagues in his own home.

Speech neuroprosthesis works by deciphering intracortical neural activity during attempted speech into the phonemes being spoken, and then assembling those phonemes into words that are shown on-screen in real time and read aloud in his own voice.

The speech neuroprosthesis worked on the first ever day of use, achieving over 99% word decoding accuracy with a 50-word vocabulary.

On the second day, researchers expanded the vocabulary to over 125,000 words and still achieved over 90% decoding accuracy.

GitHub.
Video.

Google will have fully self-designed Tensor G5 chips made on TSMC’s 3nm process, ending 4-years of work with Samsung, adding Tensor G5, made for Google Pixel smartphones and Gemini AI, will also use TSMC advanced packaging, InFO-POP.

The report also cites an unnamed investment banker saying Google’s Axion CPU is being made on TSMC’s 5nm process.

MindSearch: Mimicking Human Minds Elicits Deep AI Searcher

MindSearch with open-source models can already deliver a competitive solution to the proprietary AI search engine

GitHub
Paper

Nous Research introduced 𝐇𝐞𝐫𝐦𝐞𝐬 𝟑: the latest version in Hermes series, a generalist language model 𝐚𝐥𝐢𝐠𝐧𝐞𝐝 𝐭𝐨 𝐲𝐨𝐮.

Hermes 3 is available in 3 sizes, 8, 70, and 405B parameters.

Hermes has improvements across the board, but with particular capability improvements in roleplaying, agentic tasks, more reliable function calling, multi-turn chats, long context coherence and more.

Paper.

Model was trained in collaboration with Lambda, and they are now offering it for free in a chat interface here.

Airwallex, the Tencent-backed digital payments startup, just hit $500 million of annual run rate (ARR) revenue and is seeking to get IPO-ready by 2026

CEO Jack Zhang said that he wants the company to reach $1 billion of ARR by 2026 or 2027: "That is the goal."

Airwallex is experimenting with digital "AI workers" to boost sales.

HuggingFace dropped an in-depth tutorial on how to build your own robot!

Teach it new skills by showing it a few moves with just a laptop.

Then watch your homemade robot act autonomously:

1. Find tutorial through this link.

Unlock the power of end-to-end learning — like LLMs for text, but designed for robotics!

You will learn how to train a neural network to directly predict the next motor rotations straight from camera images.

2. The first guide you to HF bill of materials to order your robot parts (in $, £ or €) through this github page.

3. Also provided a guide to 3D print your parts. As soon as you gathered everything, you can start the assembly.

They've made some detailed videos to make things easy for you:

1.
2.
3.

Programmability in payment and settlement. The IMF has released a working paper on programmability in payment and settlement.

Highlights

1. #Programmability in #payments and #settlement has yet to fully realize its potential to support policy goals such as fostering #innovation, enhancing efficiency, improve safety and reduce fragmentation.

2. In the context of payments and settlements, programmability is the capability to perform #financial operations through logic implemented in #computerprograms.

3. Active experimentations, such as #CBDC and #assettokenization,  and even live implementations are emerging in the #financialsector. However, technical, regulatory, and financial risks introduced by new capabilities like #smartscontracts need to be understood and addressed.

The best Google Search engineering explainer dropped.

This was reverse-engineered from 1000s of leaked Google court documents.

No one else has a truly web scale search engine in ~25yrs. Must read for software engineers.

Roblox has 380M monthly users and makes $3.2B/yr as a $26B company.

But it’s not profitable.

The main reason is they spend 44% of revenue on R&D, more than gaming giants Unity and not far from all of PlayStation!

Roblox engineers make 2-3x for the same role as Unity / Sony.

❗️Soli: Ubiquitous Gesture Sensing with Millimeter Wave Radar.

This research shows how AI and non-visual sensors can create new ways to perceive and interact with the physical world.

Soli's big ideas:

• Pinpoint precision using millimeter-wave radar
• Compact, energy-efficient design
• Non-visual sensors can enable new modes of interaction and computing.

This pioneering work in radar-based gesture sensing paved the way for the next era of physical AI.

You can build an AI voice agent. It takes <30 minutes and 2 tools.

how to do this?

1. If you want to go the extra mile in believability (or use the agent for other things) - clone your own voice.

You can use ElevenLabs, and provided ~1 minute of audio of myself speaking as training data.

2. Connect your Eleven Labs voice into Vapi AI.

Once you create a Vapi account, go to Provider Credentials ➡️ enter your Eleven API key ➡️ enter your Eleven voice ID.

You should then be able to build an Assistant using your Eleven voice.

3. Spin up an assistant

You can use GPT-4 Turbo as a base and plugged in your Eleven Labs voice.

Then comes the prompting. You give it the basics (application purpose, phone number, date of birth, + when to schedule).

There's lots more optimization possible.

4. Buy a phone number + trigger a call.

You can use Vapi to get a Twilio number (for $2 / month) - you can pick the zip code.

5. Evaluate the results.

Via Vapi's call logs, you can view the trannoscript and listen to the call record post-hangup.

Amazon plans to acquire AI chipmaker and model specialist Perceive for $80 million in cash to boost the company’s large language models and edge computing capabilities.

Microsoft has released the Phi-3.5 series of models!

Phi-3.5-MoE-instruct: 42B param MoE, 6.6B active params, 128k context length, trained on 4.9T tokens on 512 H100s, multilingual (10% of dataset).

Phi-3.5-mini-instruct: 3.8B param dense, 128k context length, trained on 3.4T tokens on 512 H100s, multilingual.

Phi-3.5-vision-instruct: 4.2B param with image encoder, connector, projector, and Phi-3 Mini LM, 128k context length, trained on 500B vision+text tokens on 256 A100s, enables multi-frame image understanding and reasoning

NATO released a research describing the role of #AI models in #digitaladvertising, highlighting their use in targeted #persuasion

AI in digital marketing promises significant benefits in terms of #hyperautomation at scale and #personalization, however malicious players could leverage the same potential to deploy manipulation initiatives and precision persuasion campaigns.

Big crypto payment news: MetaMask, Mastercard, and Baanx have launched the pilot of MetaMask Card - the world's first Mastercard payment card that enables direct spending from your MetaMask wallet.

Eligible users can now make everyday purchases with their crypto anywhere Mastercard is accepted.

The pilot phase is kicking off in the EU and the UK, offering a few thousand users the chance to sign up for a MetaMask Card, which includes integration with Apple Pay or Google Pay for immediate use. Eligible currencies at launch include USDC, USDT, and WETH on Linea.