AI & Startups March 4 - March 10

Gen AI Product News

OpenAI ChatGPT quietly rolls out ‘Read Aloud’

A new ChatGPT feature called ‘Read Aloud‘ has been quietly rolling out to a select group of users on both mobile and desktop versions, with no official announcement from OpenAI.

The details:

• Read Aloud allows ChatGPT to read the answers it generates aloud using an AI-generated voice.
• Users can select and change the voice of ChatGPT in settings, with five options currently available: Ember, Breeze, Cove, Juniper, and Sky.
• The rollout of Read Aloud appears to be limited to a small number of users, but we expect an official announcement and broader rollout shortly.
• Images were obtained via direct message to our team from the user @TeddyNAE on X.

Anthropic releases Claude 3 and new benchmarks

Anthropic just announced the release of its next-generation Claude 3 model family, with the top-tier ‘Opus’ version outperforming top models like GPT-4 and Gemini Ultra across major benchmarks.

The details:

• The Claude 3 lineup consists of Opus, Sonnet, and Haiku, each offering different levels of capability, speed, and cost.
• Opus outperforms GPT-4 and Gemini Ultra on benchmarks, including expert knowledge, reasoning, coding, and mathematics.
• Claude 3 also claims new vision capabilities on par with leading models like GPT-4V.
• Opus displays a near-perfect 99% recall ability on processing long context prompts.
• All three models boast a 200k context window (with the ability to handle as high as 1M).
• Claude has removed the waitlist for API usage, allowing builders to immediately integrate the enhanced new models.

AI In geopolitics

U.S Army explores GPT-4 as a battle adviser

The US Army Research Lab is reportedly testing commercial AI chatbots like GPT-4 as battlefield planning assistants in war game simulations.

The details:

• In an experiment, AI chatbots proposed battle plans for a war scenario in the game StarCraft II.
• OpenAI's GPT-4 outperformed other AI models tested, though they were still imperfect.
• The DoD has identified a whopping 180 possible military uses for AI, working with companies like Palantir and Scale AI to integrate the tech.
• Experts caution against using AI for military decisions, citing feasibility and ethical/bias concerns.

AI impact on jobs

?AI replaces 700 customer service reps Klarna

Swedish Fintech giant Klarna released a blog praising the success of its OpenAI-powered customer service chatbot, which handled a whopping 2.3M conversations in the last month.

The details:

• Klarna estimates the assistant will drive $40M in additional profit in 2024.
• The chatbot finished the equivalent work of 700 full-time employees.
• The average time to resolution dropped from 11 minutes to 2 minutes, with the bot available 24/7 and able to converse in over 35 languages.
• The Swedish firm laid off 700 employees in 2022 and said it will no longer be hiring new staff outside of engineering due to AI.

Inflection AI upgrade nears GPT-4

Inflection just launched Inflection-2.5, a major upgrade to the model powering its Pi personal AI assistant — now capable of delivering near GPT-4 level performance.

The details:

• Inflection-2.5 achieves 94% of GPT-4's performance across tasks, despite using just 40% of the training compute.
• The upgrade displays major increases from Inflection-1 across the board, particularly in coding and mathematics.
• Inflection noted major gains in user engagement, boasting 6M monthly users and an average chat length of 33 minutes.
• The chatbot also now incorporates real-time web search

Gen AI VC

Andreessen Horowitz Eyes $6.9 Billion for AI and More in Massive Fundraising Push

Paris-based Zama raises €67 million Series A to build privacy-preserving blockchain and AI

OVERJET Raises $53 Million: The Largest Investment in the History of Dental AI

Metaplane nets $13M to detect data anomalies with AI

NLX Raises $12 Million in Funding: How Airlines Use Its AI Technology to Enhance Their Operations

Theia Insights raises $6.5m to revolutionize global investment with AI

亚马逊 Backs Glacier ’s $7.7 Million Capital Raise to Boost Recycling with AI and Robots

Singapore's Ai Palette secures $5.8M in Series A1 funding led by Tin Men Capital

London-based Fluent raises €6.9 million to make its AI-powered data analyst sweep across the workload

Propensely Inc. lands $3M seed to propel AI-powered platform for B2B cross-selling

Singapore's Temasek in discussions to invest in OpenAI

Global VC Funding Settles Around $20B In February But AI Share Increases

AI Startup Raisings Jump 40% in Q1: How To Ride This Wave?

AI2 Incubator secures $200M in AI compute resources for portfolio companies

Jurny Launches New Crowdfunding Round to Fuel its Next Generation AI

AI Research

Trust in AI tumbles in new study

Trust in AI tool companies has dropped significantly worldwide in recent years, according to a newly released 2024 Trust Barometer study.

The details:

• Responses show a stark 26-point gap between high trust in the overall tech industry (76%) and much lower trust in AI specifically (50%).
• AI trust in the U.S. plummeted by 15 points from 2019 to just 35%, while global numbers fell 8 points to 53%.
• AI concerns centered on privacy, devaluing humanity, societal h

Researchers create self-spreading AI malware

Cybersecurity researchers just developed the first known computer ‘worm’ designed to target and spread through interconnected AI ecosystems.

The details:

• Morris II uses adversarial prompts to trick AI models into replicating malicious inputs/actions and spreading them to other AI agents.
• Researchers successfully used the worm in experiments to steal sensitive data and send spam via AI email assistants.
• The malware worked against OpenAI, Google, and open-source AI models in experiments.
• It can also embed malicious prompts inside images to infect AI systems.

AI & EDUCATION

A new study in Ghana found that students who used Rori, an AI-powered chatbot math tutor, achieved significantly higher math scores compared to just regular math lessons.

The details:

• Oxford researchers divided 1,000 students into two groups, with one using Rori for two sessions per week in addition to normal math lessons.
• After 8 months, the Rori students had significantly higher math growth scores than the control group.
• Rori works on mobile devices and low-bandwidth networks.
• The projected cost per student is also just $10.

Kerala School Introduces Iris, India's First Generative AI Humanoid Teacher

Emerging Markets AI

India announces $1.2B investment in AI projects