ai

2026-01-08 Daily Ai News

Published: (January 8, 2026 at 06:05 PM EST)
3 min read
Source: Dev.to

Source: Dev.to

AI in Healthcare

OpenAI’s ChatGPT Health is integrating electronic health records through partnerships with b.well, Apple Health, MyFitnessPal, and Peloton. The service offers personalized explanations of lab results, doctor‑visit preparation, insurance comparisons, and longitudinal pattern detection. To address privacy concerns, these capabilities are siloed from general chat functions.

  • Collaboration spans 260+ physicians and targets the 230 million weekly health queries (about 5 % of global AI prompts, roughly 40 million daily users).
  • The rollout is currently U.S.-only and tied to iOS, which introduces friction for broader adoption.

AI in Finance

JP Morgan has eliminated external proxy advisors, replacing them with Proxy IQ, an in‑house platform that parses annual meeting data to guide portfolio decisions.

Market Share and Competition

  • ChatGPT’s market share fell 22 percentage points over the past year, dropping from 86.7 % to 64.5 %.
  • Google Gemini surged to 21.5 %, driven by Android ubiquity and app integrations such as YouTube summaries, which have shifted about 40 % of users from ChatGPT.
  • xAI’s Grok tripled to 3.4 % among X’s 250 M daily users, backed by a $20 B funding round that makes it the second‑most‑funded lab after Anthropic.
  • xAI emphasizes real‑time X data and a “maximum truth‑seeking” ethos, prioritizing attention over raw parameter counts.
  • Google’s arsenal includes TPUs (since 2015), the original Transformer architecture, DeepMind acquisition ($400 M), and a $2.7 B repurchase of Noam Shazeer’s stake, revitalized under Sergey Brin’s return and Gemini’s cross‑app push.

OpenAI is losing roughly 20 % of its share annually, highlighting distribution as a new competitive moat. Elon Musk has claimed that “within 3 years all knowledge work WILL be done by AI,” a statement framed by some as central‑bank‑fueled hype.

Open‑Source and Hardware Developments

  • South Korea’s state‑backed AI trio now leads Hugging Face trends, showing that national open‑source initiatives can challenge the US‑China duopoly.

Andrej Karpathy’s NanoChat

Karpathy released NanoChat v1, a series of experiments validating Chinchilla‑optimal regimes at toy scale:

# scaling_laws.sh
python train.py --model d10 --batches 0.5M --gpus 8xH100 --no-accumulation
# miniseries.sh
python evaluate.py --baseline GPT-2 --target GPT-3 --metrics CORE
  • Trains 0.5 M batches on 8 × H100 GPUs for ~4 hours at a cost of ≈ $100, achieving N/D exponents of ~0.5 with an 8‑token horizon constant.

NVIDIA Hardware Roadmap

Jensen Huang announced 10× efficiency gains moving from Hopper → Blackwell → Rubin, enabling:

  • AI‑powered factories
  • Omniverse simulations on RTX Pro
  • Thor/Orin “brain” chips with safety‑focused OS

The focus is shifting toward vertical applications (e.g., EMS assembly, surgical robots) rather than purely horizontal generality.

  • LTX‑2 releases full open‑weights, training code, and distilled models for local RTX inference, featuring native 4K audio‑video sync, stable lip‑sync, and a LoRA‑fine‑tunable “scientist mode” (hypothesis‑evidence‑conflict‑revise loops).
  • MiroThinker 1.5‑30B “search agent” outperforms a 1 T‑parameter baseline on the BrowseComp‑ZH benchmark (66.8 % vs. Kimi‑K2’s 62.3 % at 30× the parameters/cost).
  • Reachy Mini saw a preorder surge at CES, promising ~90‑day delivery for builder‑focused bots.
  • “Everything is a file” agent hacks (e.g., grepping directories for dynamic context) are gaining traction.

Claude Code and Cloud Independence

Claude Code’s 30‑minute “right way” tutorials and recent Codex data‑exploration wins are reducing reliance on cloud services, though cybersecurity concerns remain (“calm before the storm”).

Global Open‑Source Landscape

China’s Qwen model now eclipses all US/EU downloads and fine‑tunes, according to Interconnects’ 8‑plot ecosystem tracker.

Economic and Labor Implications

  • Elon Musk predicts AI will automate 50 % of white‑collar tasks today, compressing timelines toward “all work” in five years.
  • David Shapiro argues that the moralization of work masks its role as a systemic bargaining chip, suggesting new levers such as decentralized sentiment boycotts.
  • A $100 K MRR SEO bot spends $99.5 K on API calls (OpenAI $18 K, Anthropic $17 K, Gemini $1.5 K) plus infrastructure, netting $437 profit—illustrating the thin margins and autonomy of founders.

Notable Quote

“The correct way to think about LLMs is… a family of models controlled by a single dial (the compute you wish to spend).”
— Andrej Karpathy (full post)

Back to Blog

Related posts

Read more »