Hello readers,
Welcome to the AI For All newsletter! Today, we’re talking how AI is decrypting how individual neurons encode grammar, RAG vs. fine-tuning, and more!
AI in Action: Decoding Grammar at the Neuron Level

For all we know about the brain, the question of how individual neurons produce language has remained stubbornly out of reach. We've had regional maps — which lobes light up during speech, which areas connect to comprehension — but the cellular level, where actual words and sentences emerge from biological tissue, was largely unknown. A study published in Nature begins to change that. Researchers at Massachusetts General Hospital, funded by the NIH, used AI to analyze single-neuron recordings from eight patients during natural conversation, and for the first time mapped how individual brain cells encode the structure of spoken language.
All eight patients had microelectrode arrays implanted for epilepsy monitoring — not for this research. The team recorded hours of wide-ranging conversations with each participant, then aligned transcripts of those conversations with the live activity of hundreds of neurons in the frontotemporal cortex. What they found, using natural language processing models trained on the recordings, was a clear division of labor among neurons: some handled basic word meaning and grammatical role, while others took on the more complex task of grouping phrases into structured sentences. The models could even distinguish between similar phrases in different contexts, capturing the specificity of meaning rather than just recognizing words in isolation.
The downstream implications point in two directions. One is basic science — the researchers describe these findings as “setting the table” for new questions about how speech is actually generated, at a resolution that wasn't previously possible. The other is clinical. Existing neural interfaces for people with communication disorders — locked-in syndrome, ALS, severe stroke — have made real progress in recent years, but they work at a much coarser level than individual neurons. Knowing which specific cells encode grammar, meaning, and context opens a path toward devices that can decode what a person actually intends to say.
🔥 Rapid Fire
Exclusive: OpenAI lost $38.5 billion in 2025 per leaked financials
OpenAI burned $3.7 billion in first three months of 2026
Addendum: The Information’s headline leaves out that OpenAI spent $8.6 billion on “research and development” in Q1 2026
$8.6 billion is almost half of its entire 2025 R&D spend ($19.18 billion)
Analysis: The Silicon Valley Bubble (Part 1)
NVIDIA turns to debt markets for first time in five years — June 15, 2026
“If NVIDIA wants to continue growing its revenues at the current pace, the company will have to dig deep into its coffers. I would not be shocked if NVIDIA surprises the market and tries to raise significant amounts of debt via bond issuances.” — JustDario on May 21, 2026
Executives and employees are struggling with Meta’s chaotic AI strategy
Meta plans to curb employees’ AI usage as token costs reach billions
Banks run up huge bills from AI experiments
Businesses are becoming increasingly conscious of costs
Anthropic sued over limits on its most expensive plans
The federal lawsuit alleges that Anthropic oversold usage allowances
Everything GTM. One platform.
Small teams don't have time to stitch together five tools and hope it works.
Apollo gives you everything you need to find leads, reach them, and close deals — all in one place:
230M+ verified contacts
AI-powered outreach
Data enrichment
Inbound lead capture
Meeting scheduler
And more
Stop juggling tools and start building pipeline that scales.
With Apollo, the AI revenue engine powering 4M+ users.
📖 What We’re Reading

Many companies face a crucial architectural decision that directly impacts cost, accuracy, scalability, and operational risk. The question is whether they should use retrieval-augmented generation (RAG), fine-tuning, or a combination of both. Those who make the right decision often find an easier path to success. On the other hand, those who get it wrong typically end up with brittle systems, runaway costs, and outputs they can’t trust.



