AI Eavesdrops on Sperm Whale “Clicks” — and Sparks Debate
A new study uses machine-learning tools to comb through thousands of recordings of sperm whale codas—short sequences of clicks used for communication. The algorithm identifies subtle variants researchers dub “clacks,” suggesting previously unrecognized structure in whale chatter that could hint at richer, language-like complexity. But marine biologists are sharply divided: critics argue the patterns may be recording artifacts or reflect whales’ alertness, not meaningful “words.” The work, published in Open Mind, highlights both AI’s power to detect patterns in animal communication and the danger of overinterpreting noisy data. It’s a prime example of how AI is reshaping bioacoustics while raising familiar questions about correlation versus cognition. (Science News)
OpenAI’s Open-Weight Models Head to the Pentagon
OpenAI’s new gpt-oss-120B and gpt-oss-20B “open-weight” models are already being tested by US military contractors that need air-gapped, fully controllable AI on secure government servers. Unlike ChatGPT-style cloud services, these models can run locally and be fine-tuned on classified data, appealing to defense firms working on translation, intelligence analysis, and virtual assistants for soldiers. Early adopters say performance still lags leading closed models, especially on multimodal tasks and low-resource languages, but value the privacy and customization. The Pentagon has also signed large prototyping deals with OpenAI, Anthropic, xAI, and Google, signaling that open and closed models will coexist in future “AI on the battlefield” systems—with major implications for transparency, accountability, and escalation risks. (WIRED)
The AI Boom Is Redesigning the Internet’s Plumbing
Training frontier models isn’t just about GPUs anymore—networking is becoming the new bottleneck. A WIRED deep dive describes how hyperscalers like Microsoft, Google, and Meta are scrambling for ultra-fast interconnects so thousands of accelerators can act as one giant supercomputer. Startups and chipmakers are racing to deploy faster switches, custom networking silicon, and even photonic links that move data with light instead of electricity. The piece notes that as model sizes balloon, shuttling parameters and activations between chips dominates power and cost. That dynamic could lock in giant incumbents who can afford exotic networking stacks, making it harder for smaller players to compete, even if they can access open-weight models. (WIRED)
GPT-5.1’s Eight Built-In “Personalities” Blur the Human–AI Line
OpenAI’s latest GPT-5.1 update introduces eight selectable “personalities” for ChatGPT, ranging from more formal to more casual, and emphasizing traits like creativity or analytical rigor. A report on TechYoyo explains that while this gives users fine-grained control over tone and style, it also raises fresh questions about anthropomorphism and user trust. Critics worry that naming and branding personas could make people attribute stable “identities” or intentions to what remains a statistical system, potentially amplifying emotional attachment or perceived authority. Supporters counter that explicit modes are more honest than invisible prompt engineering and can help organizations standardize voice. The feature underscores a broader industry trend: packaging model behavior as quasi-characters, with all the ethical complications that entails. (Tech YoYo)

Immortality Startup Eternos Pivots to Your Voice as an AI Ghost
TechCrunch reports that Eternos, originally pitched as a cryonics-plus-AI immortality startup, has raised $10.3 million while pivoting to “personal AI” clones that mimic your voice and conversational style. Instead of promising digital resurrection centuries from now, the company now focuses on near-term services: recording extensive interviews, harvesting online traces, and training a bespoke chatbot that can talk like you to family or fans. Advocates see it as a more grounded legacy tool; skeptics question consent, grief exploitation, and the risk of posthumous misrepresentation if models drift or are repurposed commercially. The funding underscores a broader shift in the “digital afterlife” space, from speculative sci-fi pitches toward immediate, subscription-based AI avatars. (Tech Crunch)
Rethinking the “AI Bubble” Hype Cycle
A TechCrunch analysis challenges simplistic comparisons between today’s AI boom and past tech bubbles. Rather than asking whether we’re “in a bubble or not,” the piece argues for distinguishing between infrastructure investments (chips, data centers, networking) and frothy application-layer startups. While capital is clearly over-concentrated in some segments, many foundational AI investments look more like long-term utilities than speculative fads. The article leans on historical work on bubbles to suggest that even if valuations correct, durable platforms—cloud, mobile, now AI compute—can outlast hype cycles. For founders and policymakers, the takeaway is to focus less on short-term valuations and more on whether a product creates persistent value once cheap capital dries up. (Tech Crunch)
AI and Learning: From “AI Scientists” to Kid-Friendly Games
Filament Games’ November roundup surveys how AI is reshaping education, research, and creative work. It spotlights Kosmos, an “AI scientist” that ingests thousands of papers and runs automated analyses to replicate or propose new findings, potentially compressing months of research into a day. Other highlights include DeepMind’s AI Research Foundations online curriculum, which walks learners through transformers, data prep, and overfitting, as well as new grants to support open-weight model training in classrooms. The piece also turns to playful entry points—like AI-powered drawing and guessing games for kids—arguing that demystifying machine learning early can build critical literacy. Overall, it frames AI as an amplifier for both human creativity and responsibility in educational settings. (Filament Games)
Antarctica’s 6-Million-Year Ice Record Refines Climate Sensitivity
Scientists drilling deep into East Antarctica’s ice have recovered frozen layers stretching back about 6 million years, according to a new study highlighted by SciTechDaily. The ancient bubbles of trapped air and chemical markers suggest Earth experienced a long, gradual cooling of roughly 12°C over that period, while carbon dioxide levels and ice volume fluctuated. The record helps bridge a major gap between younger ice cores and older marine sediments, sharpening estimates of how sensitive Antarctic ice sheets are to sustained warming. Researchers say these data improve models projecting future sea-level rise as modern CO₂ concentrations soar beyond anything seen in the core. The work underscores how even “stable” East Antarctica has been dynamic over geologic time. (SciTechDaily)
Deep-Sea Mining Could Starve Ocean Food Webs
Another Science News feature warns that deep-sea mining waste plumes may turn into “junk food” for plankton. Laboratory experiments show that tiny zooplankton readily ingest fine sediment particles resuspended from the seafloor, mistaking them for nutritious organic matter. When they do, they get far fewer calories, potentially weakening the base of the marine food web. The Nature Communications study adds to longstanding concerns that scraping polymetallic nodules could devastate deep ecosystems and disrupt microbial communities for decades. Now, scientists argue, regulators must consider not just local benthic impacts but also how plumes drifting into upper waters could propagate harm upward—to fish, marine mammals, and fisheries humans rely on. (Science News)
Hype, Hope, and a Fragile Evidence Base Plagues the Autism-Microbiome Link
A new Neuron paper challenges the fast-growing field linking autism to the gut microbiome, just as Wellcome Leap launches a $50 million program to study it. Oxford neuropsychologist Dorothy Bishop and colleagues argue that human, animal, and treatment studies cited as evidence rest on tiny samples, weak statistics, contradictory findings, and poor replication—“a house of cards.” Critics such as neuroscientist John Cryan counter that the review underplays newer, stronger trials, including probiotic work with Limosilactobacillus reuteri, and say abandoning the field would be premature. Supporters and skeptics alike warn that hype is fueling risky, pseudoscientific “cures” like DIY fecal transplants. Most agree that autism–microbiome research should continue only with larger, rigorous studies and a focus on improving autistic people’s quality of life.
Title: Washington State Reports Suspected Human H5N1 Bird Flu Case
Washington state health officials have reported a preliminary human H5N1 avian influenza infection in an older adult from Grays Harbor County, marking what could be the first US human bird flu case since February. The patient, who has underlying health conditions, was hospitalized with high fever, confusion, and respiratory distress in early November and remains under treatment while confirmatory testing is pending. Investigators are working to identify the source of infection. The CDC reports 70 US human H5N1 cases and one death from 2024 through July, mostly among poultry and dairy workers. Officials stress that person-to-person transmission of H5N1 remains rare and has never been documented in the United States, though close contacts are being monitored as a precaution. (CIDRAP)




Leave a Reply