Saturday Quotes: The sound of music, sneaky birds, better training for LLMs. Plus: Diversity improves research

In the small fishing village where I grew up, we didn’t have much. But we helped our neighbors, raised our children to respect the sea, and embraced an inclusive scientific methodology with a cross-section of gender, race, and gender among study participants that enriched the results of our research.

This week we reported on a study that turned previous results of brain research on their head by including both men and women. Scientists also report that cuckoos may have a more important ecological function than previously thought. Plus a new training modality for LLMs and insights into how people distinguish music from speech.

Jerkbirds increase biodiversity

Many cuckoos are brood parasites and lay their eggs in the nests of other birds; When they hatch, the cuckoo young kick the other eggs out of the nest so that their foster parents are not distracted by their own young, leaving them to feed their baby parasite for weeks when they could have been raising their own family.

Songbirds have adapted to reject strange-looking offspring, so the bronze cuckoo’s nestlings have evolved to resemble their host parents. A new study by researchers at the University of Cambridge reports that this co-evolution is driving the emergence of new cuckoo species and thus biodiversity in general.

The researchers conducted a large-scale analysis of all cuckoo species and found that lineages that are most costly to their hosts in terms of resources and rearing time have adapted to be most similar to their host birds. Dr. Clare Holleley from CSIRO said: “This finding is significant in evolutionary biology and shows that co-evolution between interacting species increases biodiversity by driving speciation.”

Involving women is useful, scientists believe

In the field of gender diversity, much scientific research is stuck in the 1950s, with researchers primarily recruiting men as subjects for studies and surveys under the assumption that the findings apply equally to women. Now, researchers at Weill Cornell Medicine report the first evidence that astrocyte receptors in the brain create opposing effects on cognitive function between male and female clinical models, suggesting that astrocytes contribute to sex-specific brain mechanisms. Previous studies of astrocyte cognitive function only included males, leading to the widespread assumption that the mechanism was the same across genders.

Gender differences are known in many neurological disorders, including schizophrenia, stroke and dementia. Focusing on mGluR3, a glutamate receptor in astrocytes and a key altered gene in dementia, the research team selectively manipulated astrocytes in animal models to investigate the effects of mGluR3 on learning, memory, and cognitive and behavioral outcomes.

In women, increasing mGluR3 levels improved memory in older women, and decreasing them decreased memory in younger women. In men, reducing mGluR3 improved memory and increasing its levels had no effect. “Therapeutics that affect astrocytic receptors may induce sex-specific cognitive effects in part due to the divergent roles of astrocytes in men and women,” said Dr. Anna Orr.

Better pedagogy for LLMs

By training LLMs with a new modality that is similar to language training in the human brain, researchers at Hong Kong Polytechnic University have developed AI models that are more similar to humans. Current LLM training is designed around contextual word prediction. Instead, the researchers examined next-sentence prediction tasks, simulating discourse-level comprehension in the human brain to evaluate the coherence of a few sentences.

They trained two models, one with NSP enhancement and one without. They were also trained in standard word prediction. The researchers compared patterns from the models with fMRI data collected from people reading connected or disconnected sentences. The model with NSP training better matched human brain activity than the LLM without NSP training. The researchers hope to create new training modalities for LLMs that don’t rely solely on a huge corpus of text to achieve intelligent results.

Music, speech distinguish

People are able to distinguish the sounds of music from speech and concentrate closely on speech in an environment with music. An international team of researchers has mapped the process in a study that could improve therapeutic programs that use music to help people with aphasia regain the ability to speak.

The researchers played sound clips to study participants, who were told they would hear noise-masked speech or music, and asked to distinguish them. As participants sorted hundreds of clips, the researchers tracked the extent to which speed and regularity affected their judgments.

According to the researchers, the brain uses simple acoustic parameters to distinguish speech from music. They found that slower, more stable sound samples with random noise sound more like music, while fast, irregular samples sound more like speech. Specifically, speech is two to three times faster than most music. Furthermore, the changes in volume (or amplitude modulation, if you want to impress a peer reviewer) are stable in music, while the volume of human speech changes frequently.