PeopleofSuperintelligence

Researcher

Alexander A. Alemi

Also known as Alex Alemi

Alexander A. Alemi is a researcher whose work connects information theory with deep learning. After studying physics at Caltech, he completed a PhD in theoretical condensed matter physics at Cornell under Jim Sethna and worked on text segmentation, image classification, information bottlenecks, machine-learning thermodynamics, model scaling, and compressed language models. On a podcast in November 2025, he discussed the information bottleneck and scaling laws as an AI researcher at Anthropic, after earlier work at Google Brain and Disney. His current research at Anthropic focuses on scaling and the intersection of information theory and deep learning as of October 2026.

Now at
Anthropic
Links
Google Scholar, X and GitHub

Timeline

  1. 2015

    Physics and text segmentation

    At Caltech, he studies physics and completes a PhD in theoretical condensed matter physics at Cornell under Jim Sethna. With Paul Ginsparg, he publishes a 2015 method that uses semantic word embeddings for text segmentation and applies it to scholarly articles from arXiv.

  2. 2016

    Inception-ResNet-v2 at Google

    In August 2016, Google software engineer Alex Alemi announces the release of Inception-ResNet-v2, a convolutional network that achieves a new state of the art on ILSVRC image classification. Among the credited model architects is Alemi.

  3. 2017

    Deep variational information bottleneck

    Deep Variational Information Bottleneck appears at ICLR in 2017 with Alemi, Ian Fischer, Josh Dillon, and Kevin Murphy as coauthors. Using a neural network and the reparameterization trick, the method parameterizes the information bottleneck; models trained with its objective outperform other regularization methods on generalization and adversarial robustness.

  4. 2018

    Thermodynamics of machine learning

    A 2018 paper by Alemi and Ian Fischer, TherML: The Thermodynamics of Machine Learning, connects a framework for machine-learning objectives with thermodynamics.

  5. 2024

    Scaling exponents across optimizers

    Scaling Exponents Across Parameterizations and Optimizers, coauthored by Alemi in 2024, examines how parameterization and optimizer choices affect model scaling. The authors propose Adam-atan2, a scale-invariant version of Adam that eliminates its epsilon hyperparameter.

  6. 2024

    Language models on compressed text

    Large language models are the focus of his 2024 coauthored work on neurally compressed text. Equal-Info Windows segments text into blocks with equal bit length, and the method outperforms byte-level baselines on perplexity and inference speed.

  7. November 2025

    The information bottleneck podcast

    During a podcast in November 2025, he discusses the information bottleneck and scaling laws as an AI researcher at Anthropic, after earlier work at Google Brain and Disney.

  8. As of October 2026

    AI research at Anthropic

    As of October 2026, he works at Anthropic and focuses on scaling and the intersection of information theory and deep learning.

Career

  1. NowAnthropicAI researcher
  2. GoogleSoftware engineer

Education

  1. California Institute of Technology
  2. Cornell UniversityPhD

From the biography.

  1. Of 943 people shaping AI who worked at Google, 446 work there nowIncludes Alexander A. Alemi. Our estimates of where the 943 people shaping AI who worked at Google and Google DeepMind work now: 446 at Google and 452 across 278 other organizations.
  2. 22 of 133 people shaping AI with a former OpenAI role now work at AnthropicIncludes Anthropic. Our estimates of where OpenAI recruits came from and where 133 people with former OpenAI roles work now, including the companies and labs they founded.
  3. The median h-index is 48 at universities and 14 at frontier AI labsIncludes Anthropic. The median h-index is 48 at universities and 14 at frontier AI labs. We compare publishing careers, recent citations, dual appointments and job moves across 2,420 university researchers and 308 frontier-lab researchers.
  4. Companies supply the most prior-work hires at 7 of the 9 frontier AI labsIncludes Anthropic. Our estimates of the organizations where 550 people worked directly before joining 9 frontier AI labs across 620 career transitions, comparing the main feeder organizations per lab.

Sources and credits

Photo: We have no published photo of Alexander A. Alemi yet. Send us a free-licensed photo.

  1. Alexander A. Alemi alexalemi.com. Checked 21 October 2026.
  2. Text Segmentation based on Semantic Word Embeddings arxiv.org. Checked 21 October 2026.
  3. Improving Inception and Image Classification in TensorFlow research.google. Checked 21 October 2026.
  4. Deep Variational Information Bottleneck research.google. Checked 21 October 2026.
  5. TherML: The Thermodynamics of Machine Learning research.google. Checked 21 October 2026.
  6. Scaling Exponents Across Parameterizations and Optimizers arxiv.org. Checked 21 October 2026.
  7. Training LLMs over Neurally Compressed Text arxiv.org. Checked 21 October 2026.
  8. EP15: The Information Bottleneck and Scaling Laws with Alex Alemi podcasts.apple.com. Checked 21 October 2026.