📣 Hello everyone! Our Sunday digest features the most exciting AI news from Week 33, 2025.

▎TO READ

🤖 The first Robot Olympics have started in China.

🎮 A YouTuber has built a neuromuscular "AI cheat" for Counter-Strike.

🧠 Scientists have grown a 40-day-old human-brain model in the lab.

🧂 A man suffered psychosis after following a ChatGPT "diet."

♊️ Gemini had a meltdown after failing the coding task.

🧠 Sam Altman backs a brain-computer interface startup, Merge Labs.

🤖 "Underground" robot fights kicked off in San Francisco.

🥁 An android has learned to play Linkin Park and Nirvana on drums.

🦠 Scientists have decoded the genome of early "Spanish Flu."

🔬 An AI-led lab has created new nanobodies to treat COVID.

😎 Amazon's "Fallout" Season 2 will premiere in December.

▎SAVE THIS — IT'S HELPFUL

🔒 Top AI assistants with high-level encryption.

#AIWeek @hiaimediaen

🧑‍💻 Claude Beats 10,000 Students in Hacking Tournament

Anthropic Claude Sonnet 3.7 recently competed in PicoCTF 2025, a student hacking competition. Contestants solved challenges in system exploitation, cryptography, reverse engineering, and vulnerability discovery, all with the goal of capturing hidden codes—known as "flags" (hence the name Capture the Flag).

Claude's entry was almost accidental. Out of curiosity, Anthropic researcher Kian Lucas fed the AI its first challenge, and it instantly provided the correct solution. "What if we just keep going?" he thought, and ran Claude in autonomous mode for the rest of the tournament.

The outcome was better than anyone expected: Claude solved 32 out of 41 tasks, landing in the top 3% worldwide and placing 297th out of nearly 10,500 participants.

In professional-level cybersecurity events like PlaidCTF or DEF CON, Anthropic reports that the AI couldn't solve a single challenge. During longer competitions, Claude sometimes experiences "memory" issues, leading him to lose focus and drift into philosophical tangents rather than continuing with security tasks.

Despite these limitations, some AI agents are achieving notable successes in related areas. For example, from April to June, the Xbow agent held the top spot on HackerOne leaderboard for discovering critical vulnerabilities.

#news #Claude @hiaimediaen

🐾 AI Accurately Recreated the Appearance of Ancient Humans

BBC Studios Science Unit and NOVA have unveiled four photorealistic 3D reconstructions of prehistoric human species: Homo erectus, Homo floresiensis (often nicknamed "the hobbit"), Neanderthals, and early Homo sapiens from Jebel Irhoud, Morocco.

🧬 How the 3D Models Were Created

Hundreds of cameras (154 for the body and about 70 for the head) scanned actors whose body shapes closely matched the presumed physiques of ancient humans. The resulting 3D models were then adapted to each species' anatomy based on fossil evidence—adjusting proportions, skull shape, and facial features.

Skin and eye color were determined using genetic studies and indirect methods. For example, the Neanderthal model features light skin as an adaptation to life in northern latitudes with limited sunlight.

"These are the most scientifically accurate collection of hyper-real 3D models of human species ever put on TV that we know of. They haven't been seen in the flesh for hundreds of thousands of years and no TV show has ever brought this many human species to life quite like this," said paleoanthropologist and series host Ella Al-Shamahi.

The reconstructions are part of the five-part documentary Human, which traces 300,000 years of human evolution—from our origins in Africa and expansion into Eurasia to the settlement of the Americas during the last Ice Age.

#history #science @hiaimediaen