AI discovers rogue asteroid

PLUS: Meta's new Unreal Engine datasets

Good morning, human brains. Welcome back to your daily munch of AI news.

Here’s what’s on the menu today:

  • Anthropic’s newest AI model 🧠

    Claude Instant 1.2 is out. More on how it compares to its predecessor.

  • AI asteroid-discovery tool 😰☄️

    HelioLinc3D finds “potentially hazardous” near-Earth asteroid.

  • Meta’s new Unreal Engine datasets 🐘

    It released 3 “Photorealistic Unreal Graphics datasets.

APPETIZER

Claude Instant 1.2 is here 🧠

Anthropic released Claude Instant 1.2. It’s got a host of improvements over their previous model, Claude Instant 1.1.

Anthropic claims it’s better at math, coding, reasoning, and “safety.”

The claims:

1/ It scored marginally higher than 1.1 on several coding and math benchmarks.

2/ It has better quote-extraction, multilingual, and question-answering capabilities.

3/ It generates less nonsensical content.

4/ It’s more resistant to security breaches.

Our take: The changes here seem incremental, but the new model seems pretty smooth. We’re certainly not opposed to a more buttery user experience.

BUZZWORD OF THE DAY

HelioLinc3D

An AI algorithm for discovering Potentially Hazardous Asteroids (PHAs). It was specifically design for the Vera Rubin Observatory and has identified its first potentially dangerous Earth-bound asteroid.

MAIN COURSE

AI finds hazardous asteroid 🫣☄️

HelioLinc3D has found its first dangerous near-Earth asteroid. It’s an AI algorithm developed for the Vera Rubin Observatory in Chile.

Suck it, dinos.

It’s the first “potentially hazardous” asteroid discovered by AI.

HelioLinc3D is trained with data from the ATLAS survey, a program led by University of Hawaii researchers and funded by NASA.

Great balls of fire:

1/ It first discovered the asteroid, 2022 SF289, last year. It was 30 million miles from Earth — truly mind-boggling.

2/ It predicts the 600-foot asteroid will pass within 140,000 miles of Earth. For reference, the moon is 238,855 miles away.

3/ NASA-backed ATLAS missed it.

You had one job.

“Any survey will have difficulty discovering objects like 2022 SF289 that are near its sensitivity limit, but HelioLinc3D shows that it is possible to recover these faint objects as long as they are visible over several nights. This in effect gives us a ‘bigger, better’ telescope.‘”

-Larry Denneau, Lead ATLAS astronomer

Our take: more ammo for Marc Andreesen’s “AI will save the world” case. It’s nice to take a quick break from the Skynet & Terminator memes.

A LITTLE SOMETHING EXTRA

Meta’s new Unreal Engine datasets 🐘

Meta released 3 synthetically generated datasets using Unreal Engine. They’re called PUG: Animals, PUG: ImageNET, and PUG: SPAR.

PUG stands for “Photorealistic Unreal Graphics.”

The gist:

1/ Meta developed synthetic image datasets using Unreal Engine to render photorealistic image data.

2/ The datasets revealed posing and lighting limitations in ImageNET.

3/ These datasets allow researchers to study how different vision-language models capture the relationship between image and text.

Our take: Factors like lighting and viewpoint are notoriously difficult to control in real-world datasets. Perhaps these PUG datasets will make it easier for future researchers to create realistic 3D assets from AI.

MEMES FOR DESSERT

YOUR DAILY MUNCH

Think Pieces

Joe Biden signs an executive order that bans US investments in Chinese tech, including AI. It restricts investing in Chinese chips, AI tech, semiconductors, quantum computing tech, and more.

The Pope’s warning about AI: “disruptive possibilities and ambivalent effects.“

Is GPT-4 better at operating systems, databases, and creativity than other LLMs? According to a study, it’s better than 25 other models.

Startup News

Amazon tests new AI tools for sellers. The first AI tool outside its cloud services, it writes product descriptions, titles, and more.

Virtualitics secures a $37 million investment. The data visualization startup secured their Series C round, led by Smith Point Capital with investments from Citi and The Hillman Company.

Research

SlimBot Challenge on Embodied AI — a paper on a competition between university teams to create the most versatile conversational AI agents in a simulated environment.

A paper on Fan (Follow Anything) — a robotics method to streamline open-set detection, tracking, and real-time following.

A paper on FocalFormer3D — a 3D object detection model to tackle false negatives in autonomous driving scenarios.

Tools

Think Evolve Consulting [Sponsored] — Data-Science-as-a-Service. Our team of experienced Data Scientists (10+ years) is on call to augment your team and ensure your business’s success.

Team Town [Sponsored] — your own in-house design team without those pesky design bottlenecks that slow you down. The team behind Staples, Built, Jeep, and more.

Raz — Get customer feedback with AI-powered forms that ask follow-up questions.

Chatmate — Create GPT-powered chatbots with vector document retrieval capabilities.

TWEET OF THE DAY

Former Microsoft researcher/xAI mathematician’s tasteful take on AI-safety.

Tag us on Twitter @BotEatBrain for a chance to be featured here tomorrow.

RECOMMENDED READING

🏥 Healthcare AI News — 5 minutes or less is all it takes to elevate your Healthcare knowledge with this expert-curated weekly AI newsletter. Stay informed and stay ahead.

👨 The Average Joe — Market insights, trends and analysis to help you become a better investor. We like their easy-to-read articles that cut right to the meaty bits.

💌 Marketing Letter — The newsletter keeping 30k+ marketers in the loop. Read by marketers who work at LinkedIn, Techcrunch, and Disney.

AI ART-SHOW

Until next time 🤖😋🧠

What'd you think of today's newsletter?

Login or Subscribe to participate in polls.