Reading up on DeepSeek

Hacker News commenters debate the sustainability of open-weights LLMs, arguing they cannot be taken away once downloaded despite potential future restrictions or discontinuation by funders.

Chinese resellers sell cheap Claude tokens via pooled accounts and fraud, harvesting user data for distillation into Chinese AI models as Anthropic alleges Alibaba illicitly extracts capabilities.

Moonshot AI releases Kimi K2.7 Code, a 1T-parameter MoE coding model that improves real-world agentic tasks while cutting thinking tokens by 30% over K2.6.

AI chatbots are threatening high-paying wealth manager roles by automating financial advice and tax preparation tasks.

A security researcher built a deliberately vulnerable Firebase-based app and spent $1,500 testing whether LLMs could exploit it; GPT-5.5 succeeded 70% of the time, while most others failed due to guardrails or misdirected focus.

Foreign investors and founders are paying up to $9,000 for curated tours of Chinese EV and AI factories, driven by fear of missing out on technological breakthroughs.

Hacker News commenters consistently report that using AI for code review and iterative refinement yields better code, but takes significantly longer than manual coding.

Anthropic and OpenAI may have reached product-market fit based on enterprise willingness to pay $200/month for tokens, though valuation and cost sustainability remain contested.

Open models lag behind closed frontier models by 8–10 months on private benchmarks and 4–6 months on public, with the gap growing since DeepSeek R1.

OpenRouter ranks live LLM usage by tokens and spend, showing DeepSeek, Anthropic, and others dominating across general, agent, and code tasks.

Simon Willison's blog post documents rapid LLM improvements over six months, using the "pelican riding a bicycle" test as a consistent benchmark to track progress.

Commenters argue local AI is already viable for many tasks and will become the norm, driven by open-weights models and privacy concerns, not just cloud convenience.

The article argues that cheap AI from Chinese labs and Western alternatives is eroding the pricing power and market share underpinning OpenAI and Anthropic's high IPO valuations.

: HN commenters analyze Gemini 3.5 Flash's likely parameter count (~250-300B total, 10-16B active) and debate whether frontier models are actually much smaller than rumored.

Graduates booed Eric Schmidt at a commencement speech, reflecting widespread public backlash against tech elites promoting AI while advocating job displacement.

Anthropic argues that US export controls on AI chips must be tightened to ensure democracies maintain a decisive lead over China's authoritarian regime by 2028.

The piece presents a methodology that maps public benchmark scores to IQ estimates across seven dimensions, enabling comparison of AI models by intelligence, speed, and cost.

DeepSeek is raising over $7 billion from investors as it begins efforts to generate revenue.

The piece argues that AI models will commoditize like past technologies, making customer relationships and workflows more valuable than owning the base model.

Chinese AI labs leverage cultural humility, student integration, and practical focus to effectively fast-follow frontier models, contrasting with US labs' political conflicts and star-scientist culture.

DeepSeek cuts V4-Pro API prices by 75% until May 5, 2026, and reduces cache-hit prices by 90%, undercutting US rivals amid geopolitical tensions over model distillation.

Open-weight models from China are commoditizing AI capability, breaking the monopoly moat of US closed labs, prompting protectionist moves that will harm long-term US competitiveness.

The article argues that the "monetizable spread" between open and closed AI models is compressing faster than the capability spread, threatening the valuations of frontier labs like OpenAI and Anthropic.

The future of AI may shift from massive datacenter buildouts to open-source models running locally on workstations as performance gaps close and remote providers raise prices.

Multiple converging transitions—knowledge diffusion, API-first business, AI-driven enterprise graphs, and post-corporate human work—reshape society around agent-mediated, bespoke, and automated systems.

The subscription model survives only for genuine utilities and continuously-fresh-context services; the middle—static SaaS and exhausted catalogs—collapses as income shocks, context saturation, and AI repricing erode its math.

AGI capable of most cognitive work could arrive by 2028–2034, but deployment will lag capability due to verification bottlenecks, uneven automation, and institutional friction.

Anthropic publicly accused three Chinese AI startups of using around 24,000 fake accounts to scrape data for training rival chatbots.

As AI models and software commoditize, moats erode for pure software/AI firms; value concentrates in compute, energy, and relationship-based assets.

Nathan Lambert argues open models will never catch up to closed frontier systems due to resource gaps, but are crucial for US AI research and policy as an engine for exploration.

Coding agents combined with reasoning LLMs have become automated scientists, enabling a golden age where all computer science problems appear tractable through vast inference compute.

Synthetic pretraining is reshaping AI by integrating designed data generation early in training, enhancing memorization, logical hardwiring, and system simulations beyond web-curated data.

DeepSeek's Engram architecture adds conditional memory via N-gram lookup tables to LLMs, improving knowledge and reasoning benchmarks by offloading static pattern reconstruction from neural computation.

Qwen dominates open-model downloads and finetunes globally, while DeepSeek leads in large-scale models and GPT-OSS is the only Western contender gaining adoption.

The 2025 LLM landscape was defined by reasoning models, coding agents, Chinese open weight models, prompt-driven image editing, CLI tools, and OpenAI losing its lead.

Meta's $2.5 billion acquisition of Singapore-based Manus, a startup with Chinese roots, could create a new pathway for Chinese AI companies to access U.S. investment.

Hacker News commenters argue OpenAI's high valuation and spending mirror historical railroad bubbles, predicting a market downturn by 2026.

Karpathy's 2025 review identifies RLVR, jagged intelligence, Cursor, Claude Code, vibe coding, and Gemini Nano Banana as paradigm-changing developments in LLMs.

TIME names Jensen Huang and other AI leaders as Person of the Year, arguing their technology reshaped global economy, geopolitics, and daily life in 2025.

A HN user proposes forbidding "I asked $AI, and it said" replies; commenters mostly oppose a ban, preferring downvotes to discourage low-effort posts.

An empirical study of over 100 trillion LLM tokens on OpenRouter reveals rising open-weight adoption, dominant roleplay and coding tasks, and a 'Glass Slipper' retention effect.

OpenAI CEO Sam Altman declared a 'code red' to accelerate ChatGPT improvements after Google's Gemini caught up and surpassed OpenAI in user perception.

Mistral released three small dense models and a large MoE model under Apache 2.0, claiming best performance-to-cost in their size categories.

Anthropic's chief scientist Jared Kaplan says humanity must decide by 2027-2030 whether to let AI systems recursively self-improve, a step that could trigger beneficial intelligence explosion or cause humans to lose control.

DeepSeek-OCR compresses image tokens using an encoder to reduce input size, enabling larger context windows and more efficient training, not traditional OCR.

As open-source LLMs commoditize due to fierce Chinese competition, companies can build defensible moats through custom RL models trained on proprietary environments using LoRA and GRPO.

AI models are commoditizing, shifting value to applications and integration over model providers, despite massive infrastructure spending.

Reinforcement learning provides far fewer bits per FLOP than pretraining until models achieve high pass rates, limiting RLVR's ability to learn new capabilities.