Skip to content
AImpact
IT EN
← Reading paths
◦

Reading path

Data Scientist

Integrate LLMs into your workflow: RAG, embeddings, benchmarks, and fine-tuning.

For data scientists who want to use language models as real engineering components. This path traces the evolution of architectures, benchmarks, and tools that matter: from evaluating open-source models to building production-ready RAG pipelines.

  1. 01

    Why it matters to you

    The GPT-3 paper introduced in-context few-shot learning as a new evaluation paradigm, setting the baseline for understanding model capabilities and scaling without task-specific fine-tuning.

    Landmark Foundation Models

    GPT-3: the paper that opens the scaling-laws era

    OpenAI publishes 'Language Models are Few-Shot Learners' and shows that at 175B parameters a model learns new tasks from a handful of examples in the prompt.
  2. 02

    Why it matters to you

    Chinchilla rewrote the scaling laws, showing that data and parameters must be co-optimized — a foundational result for anyone designing or comparing models against benchmarks.

    Landmark Foundation Models

    Chinchilla: the big models were undertrained

    DeepMind publishes the Chinchilla paper and shows that, given equal compute, smaller models trained on far more tokens beat oversized undertrained ones.
  3. 03

    Why it matters to you

    LangChain made building RAG pipelines and agents accessible, quickly becoming the go-to stack for integrating LLMs into data-intensive production applications.

    Landmark Agents

    LangChain: the framework for LLM applications is born

    Harrison Chase releases LangChain, an open-source Python library to chain LLMs with prompt templates, memory, tools and external data sources. It will become the default stack of the first LLM apps.
  4. 04

    Why it matters to you

    LLaMA opened access to foundation models for the research community — essential for reproducible fine-tuning experiments, comparative benchmarking, and controlled local deployment.

    High Open Source Models

    LLaMA: Meta opens foundation models to research

    Meta releases LLaMA in four sizes (7B, 13B, 33B, 65B), available to researchers on request. One week later, the weights leak publicly.
  5. 05

    Why it matters to you

    Gemini 1.5's 1M-token context window changed the RAG equation: less aggressive chunking, new long-document retrieval strategies, and a higher ceiling for in-context analytics.

    High Foundation Models

    Gemini 1.5 Pro: 1 million tokens in context

    Google announces Gemini 1.5 Pro: Mixture of Experts architecture, 128K standard context, 1M in preview. New benchmark: near-perfect 'needle in a haystack' retrieval over long inputs.
  6. 06

    Why it matters to you

    DeepSeek V2 proved that MoE architectures can hit top-tier performance at a fraction of active parameters — critical for anyone evaluating cost-performance tradeoffs in production.

    High Open Source Models

    DeepSeek-V2: Multi-head Latent Attention and the first highly efficient Chinese open MoE

    DeepSeek releases V2: 236B-total / 21B-active MoE with Multi-head Latent Attention (MLA), drastically cuts KV cache, slashes Chinese API prices by 90%, and ignites a price war.
  7. 07

    Why it matters to you

    DeepSeek R1 showed that advanced reasoning can be distilled into open-weight models, enabling reproducible benchmarks and fine-tuning on reasoning-heavy tasks.

    Landmark Open Source Models

    DeepSeek-R1: open reasoning matches o1 at 1/30 the cost

    Chinese startup DeepSeek releases R1, a reasoning model with MIT-licensed open weights. Performance on par with OpenAI o1, API pricing $0.55/$2.19 per 1M tokens (vs o1 $15/$60). Nasdaq AI loses $1T in two days.