Scientists at UMass Amherst taught an AI to count bees and tell honeybees from bumblebees just by listening to their buzz, so they don't need traps that harm bees.

A tiny microphone hidden in a tomato plant. A fuzzy, orange-striped bumblebee grabs a yellow flower and shakes its flying muscles so fast that the pollen pops out like salt from a salt shaker. On a computer nearby, the AI writes: "Bumblebee. Shake-the-flower buzz." Every bee flies home safe.

It listened, guessed, checked the answer, got a score, adjusted, and tried again, like learning to ride a bike. It trained on 594 recordings made over 5 years at 43 sites. When it was very sure, it was right about 93 to 97 times out of 100. It is a coach with super patient ears, not a replacement for scientists.

1. With a grown-up, sit near some flowers (or anywhere outside).
2. Close your eyes and listen for one minute.
3. Guess what made each sound: bee, bird, car, wind?
4. Open your eyes, check, keep score, and try again.

Never touch or chase a bee; watch from a few steps away. Too cold for bees? Play indoors: a family member makes sounds (crinkling paper, tapping a spoon, pouring water) while you guess with your eyes closed.

• UMass Amherst news release (October 21, 2026): Listening to the Buzz
• Phys.org (October 21, 2026): AI listens to bee buzzes to monitor activity without lethal traps
• Paper: Figueroa L. L. et al. (2026), What's all the buzz about? Passive acoustic monitoring of bees
• Open data: Dryad dataset

A volcano baked a scroll into charcoal almost 2,000 years ago, and it's been too fragile to unroll ever since. On October 8, 2026, the Vesuvius Challenge awarded its $50,000 First Letters Prize to Erwin Nieuwlaar, a software engineer from the Netherlands. He found the first readable letters ever seen inside that scroll, PHerc. 343. It's about 14 cm long and kept at the Biblioteca Nazionale di Napoli in Naples, Italy.

Here's how he did it:

• He traced one layer of the rolled papyrus inside an X-ray scan, using the challenge's open-source spiral-fitter plus his own way of snapping it tightly to the surface.

• He ran four AI ink-detection models (one is named Hecate) and averaged their answers. None of the models had been trained on this scroll, so they were "generalising across scrolls."

• Papyrologists, who are experts in reading ancient papyrus, read the Greek letters as part of the phrase "everything that happens."

• Other teams found the letters μουσικ (mousik), which may point to a word about music.

A burnt black "marshmallow log" a little shorter than a pencil. Deep inside it, glowing yellow letters light up on a computer screen without anyone unrolling a single layer. Erwin shouted, in Dutch, "Dit is toch gewoon inkt!" ("This is surely just ink!").

The ink detectives practiced like you learning to ride a bike. They made a guess ("ink here, no ink there"), got a score, adjusted, and tried again, thousands of times. They practiced on other scrolls and still recognized ink in this new one, the way you can ride at the park after learning in the driveway.

The AI pointed and said "look here." Erwin did the tracing and the thinking, and human experts did the reading. People and AI worked as a team.

1. Write a secret message on white paper with a white crayon. Press hard.

2. Roll it up and trade with a friend or family member.

3. The ink detective brushes watercolor paint over the paper, and the hidden letters appear.

4. Bonus (with a grown-up): ask an AI helper, "What clues could help a detective tell ink apart from paper?" and brainstorm together.

• Vesuvius Challenge, "$50K First Letters Prize awarded for PHerc. 343" (Oct 8, 2026): https://scrollprize.substack.com/p/50k-first-letters-prize-awarded-for

• AI Discovery News, "The First Readable Letters Return From a Scroll Vesuvius Buried in AD 79" (Oct 15, 2026): https://aidiscoverynews.com/2026/10/15/the-first-readable-letters-return-from-a-scroll-vesuvius-buried-in-ad-79/

• The Irish Times, "Scientists decipher charred ancient scrolls buried by Vesuvius" (Oct 15, 2026): https://www.irishtimes.com/science/2026/10/15/scientists-decipher-charred-ancient-scrolls-buried-by-vesuvius/

• Background: Vesuvius Challenge, "We read an entire scroll, without ever opening it" (PHerc. 1667, June 25, 2026): https://scrollprize.org/firstscroll

Pavel Rabtsevich, a hobby astronomer in Spain who doesn't own a telescope, used AI coding helpers to search public data from NASA's TESS space telescope, and found a faraway star that dims a tiny bit about every three days. That could mean a planet is passing in front of it. He checked his AI helpers' work round after round, and now TESS will look at the star again from October 31 to November 26 to test his predictions.

A star 116 light-years away blinking like a clock (dim, three days, dim, three days), in measurements taken years apart, and nobody had noticed until now.

Be a planet hunter at home. With a grown-up, shine a flashlight at a wall in a dark room (never into anyone's eyes) and slowly move a small ball through the beam. Watch the light on the wall dim as your "planet" passes its "star." Try a bigger ball to see if it dims more. Then predict your planet's schedule ("it passes every three claps") and have someone check you.

This is a planet candidate, not a confirmed planet. Rabtsevich says himself that more observations are needed, and that even if the star dims on time in November, that alone won't prove it's a planet.

• Webb Wright, "Did Claude Help an Amateur Astronomer Discover a New Planet?" Gizmodo, October 9, 2026. https://gizmodo.com/did-claude-help-an-amateur-astronomer-discover-a-new-planet-2000824393

• Laura Bennett, "Claude Code Helps Amateur Astronomer Find a Planet Candidate," AIstify, October 8, 2026 (updated October 9, 2026). https://aistify.com/claude-code-planet-candidate-tess-pavel-rabtsevich/

• MIT TESS Science Office, "Director's Discretionary Targets" (lists active program 100, Pavel Rabtsevich, "Two-minute cadence for TIC 4206066, a nearby K-dwarf with a 1.4R planet candidate"), accessed October 12, 2026. https://tess.mit.edu/science/ddt/

• "TESS schedules a follow-up for candidate signals around TIC 4206066," NeoTeo, October 9, 2026. https://www.neoteo.com/en/tess-schedules-a-follow-up-for-candidate-signals-around-tic-4206066

• Pavel Rabtsevich, "Two Transit-like Signals in TESS Photometry of the Nearby K Dwarf TIC 4206066," preprint on Zenodo, September 25, 2026 (abstract checked via index listing). https://exa.ai/library/publication/ntcj325jyww

Machines on the Atlantic sea floor that were put there to study rocks deep inside the Earth also recorded whale songs, and an AI that human experts trained found more than 6.3 million fin whale calls in about 43 years' worth of recordings.

Picture the dark floor of the Atlantic Ocean near the Azores, Madeira and the Canary Islands. Dozens of seismometers sit there quietly, feeling the planet's rumbles. Overhead, a fin whale about as long as two school buses sings notes so low that people can barely hear them. Each note shows up as a little smudge at the bottom of a "sound picture" (a spectrogram), and an AI helper circles every smudge: more than six million of them.

"Sea-floor ears" is our show's nickname for ocean-bottom seismometers, and the host says so on air. It is not the scientists' term.

• 46 ocean-bottom seismometer stations from two networks (São Jorge and UPFLOW) in the Azores–Madeira–Canaries region of the North Atlantic. The recordings run from June 2021 to February 2023.

• 378,912 hours of recordings in total, which is about 43 years if played back to back.

• The AI found 6,291,626 fin whale calls (about 6.3 million). The authors call it the largest fin whale call catalogue to date.

• Human experts hand-marked nearly 40,000 calls to train the AI.

• About 97% precision overall (94% at São Jorge, 97% at UPFLOW). The AI also missed many quieter calls (recall was about 33% at UPFLOW), so the true number of calls is higher.

• Processing one day of recordings from one station took about 15–20 seconds.

• Calling peaked from October to February. There were three recurring song "beats" (gaps of about 12, 16 and 20 seconds between notes), and the most common one shifted during the season.

• The authors describe a possible southward movement of whales as speculative.

• Fin whales are the second-largest whale species, 75–85 feet long (NOAA).

• Japnanto, J. et al. "Large-scale bioacoustic detection using semantic segmentation: a deep learning framework applied to fin whale calls in ocean-bottom seismometer recordings." arXiv preprint 2609.13281. v1 was posted September 8, 2026 and v2 on September 29, 2026. It has not been peer-reviewed. Authors are from University College London, the University of the Azores (OKEANOS), CSIC Barcelona and the University of St Andrews.

  • Abstract: https://arxiv.org/abs/2609.13281

  • Full text: https://arxiv.org/html/2609.13281v1

• NOAA Fisheries, "Fin Whale" species page (size, second-largest whale, U.S. endangered listing): https://www.fisheries.noaa.gov/species/fin-whale

• Background on related work (a different team using AI to detect Bryde's whale calls in seismic data, June 2026): https://www.seismosoc.org/news/ai-model-hears-whale-calls-in-seismic-data/

• Optional mission tool for families: Chrome Music Lab Spectrogram: https://musiclab.chromeexperiments.com/Spectrogram/

The AI did the listening, but people taught it, checked its work, and openly said what they still don't know. Kids can use AI the same way. Don't ask "Write my whale report." Ask "Quiz me about fin whales."

1. Draw an arrow across your paper from left to right. That's time.

2. Low sounds go near the bottom of the page and high sounds near the top.

3. Sit still for one minute and draw every sound you hear. A humming fridge might be a long low line, and a bird tweet might be a tiny high dot.

4. Does any sound repeat, like a whale's beat? Count it!

5. Bonus, with a grown-up: Ask an AI helper: "I drew a sound picture. Don't fix it for me. Ask me three questions to help me make it better." Families can also try the free Chrome Music Lab Spectrogram to see their own voices as a real sound picture. It needs microphone permission, so a grown-up should turn it on.

A humanoid robot walks into a living room it has never seen — toys on the floor, a sock on the couch, a stuffed dinosaur under the table — and gets to work with its whole body. This episode of Generative AI for Kids follows Helix 2.5, Figure's newest neural network (announced September 17, 2026). After pretraining on Index, Figure's huge dataset of people showing how they move and do chores, the robot tried three jobs in homes it had never practiced in.

The wow picture: They took Helix 2.5 into 30 Bay Area homes with zero data collected in any of them. No fine-tuning. Unseen toys, towels, and beds. Tidy living rooms (toys into a basket). Fold towels into a basket. Make beds (pillows plus comforter). In a blind test with the same task data, success jumped from 9% without Index pretraining to 56% with it — over 6× better. Full-task only. No partial credit. The robot even self-corrects: steps back, repositions, walks around the bed to fix a fold. Coach, not copy machine. Not "robotics is solved" — first strong evidence that whole-body skills can transfer from human experience to new homes.

Sources:

• Figure AI, Sept 17, 2026: https://www.figure.ai/news/helix-2-5-zero-shot-30-home-generalization

Pick five toys in a room that is not your usual tidy spot. Put them into a basket, one by one. Or fold one towel carefully, corners together. If a fold is messy, step back, walk around, and try again. You just did Helix thinking. Learn from watching. Then help in a new place.

A robot arm sits on a messy desk. Blocks, a water bottle, a backpack — and a bright toy past the clutter. Pretty close is not good enough: a path that almost misses still knocks things over. This episode of Generative AI for Kids follows HardFlow, a new method from MIT researchers (published September 14, 2026) that helps generative AI invent clever paths while obeying nonnegotiable safety rules.

The wow picture: Older helpers often force every tiny sketch step to stay perfect, which can make paths shy and slow. HardFlow gives the model more freedom while it explores, then enforces hard constraints on the final answer — no collisions, plus better quality (like the shortest legal path). In tests on robotic arms, mazes, and image editing, it hit perfect constraint satisfaction and found better solutions than other methods. Humans write the rules. The AI searches. Coach, not copy machine.

Sources:

• MIT News, Sept 14, 2026: https://news.mit.edu/2026/new-method-enables-ai-safety-critical-situations-0914

Clear a table corner. Place three obstacles and one sticker prize. Draw a top view. Invent three paths for an imaginary robot finger. Cross out any that touch obstacles. Circle the shortest safe path. You just did HardFlow thinking.

Draft audio for this week's show. Not published yet.

After a year stuck, an AI coach called Get Physics Done found a Sun-dip path to Alpha Centauri in about a week. Slow down, swing closer than Mercury, fire near the Sun so panels stay small.

Kid Mission: Near-Star Map Challenge. Draw Earth, Sun, Alpha Centauri. Invent three paths. Pick one and say why it saves power or weight.

https://www.technologyreview.com/2026/09/01/1143247/ai-interstellar-journey-alpha-centauri/