Talk:Timeline of machine learning

Article is very heavily weighted toward neural networks

Neural networks are a small part of the much larger field of machine learning, yet 80% of the events in the timeline involve advances in neural networks.

Ihearthonduras (talk) 16:42, 9 February 2018 (UTC)

How to decide what goes in the table?

How are the recent events added to the page on the same scale as the historical ones? Facebook "detailing" an **internal** tool (FBLearner Flow) is an event now? or Google "detailing their work" on a different internal proprietary product is an event worth marking in the timeline of machine learning how? Why is the release of TensorFlow an event, when Theano isn't, while TensorFlow was more or less a copy of Theano at the time of release? The fact that humans have been outperformed in object recognition is not an event, but the fact that Facebook beat a benchmark on identifying faces is?

Most of the events before 2000 are actually worth mentioning and everyone knows about them. However, what has been added to that table after that period can only be described as a mess produced by companies who think they can write history by literally editing Wikipedia and mark their events important.

Is there any sort of standard to the table? Can I add anything I want as long as I have a news coverage to cite? Or perhaps do I need to be a big multi-national first?

Please sign your posts using "--" followed by four tildes. I have removed TensorFlow and FBLearner Flow per your logic. What else needs to be removed and why? Also, what needs to be added? --Hyperforin (talk) 21:45, 19 November 2016 (UTC)

"What needs to be removed and why?"

My belief is that Facebook's "Leap in Face Recognition" should be removed. There is no indication why the task of face recognition should be regarded higher than regular object recognition, for instance. I would replace this event with Krizvesky et al whose paper was the first to convince the community about the potential of deep learning. Facebook's achievement was noteworthy but it pales in comparison to the above mentioned paper and the number of citations shows it as well. Further, if Krizvesky is added, there is no need to have both, as object recognition and face recognition are very similar tasks.

Secondly, the "Sibyl" entry is only about Google "detailing" their work on an internal system via a PowerPoint presentation and some news article providing additional vague information about it. Is it necessary to form an argument for why this is not on par with the other entries? There's nothing released--nothing of impact on the machine learning community--it's just a closed-source cluster packaging previous research into a system. All the other entries have had an impact on machine learning research community, while this one is the type of news that had an impact of 60 points on Hacker News https://news.ycombinator.com/item?id=8249752

--109.151.246.114 (talk) 15:07, 20 July 2017 (UTC)

Agreed. I may remove. Looking back at the last 10 years it's clear what were the big events, and this was not one. WillSmith (talk) 21:05, 12 September 2023 (UTC)

Krizvesky's Alexnet missing from this list jeopardizes it's credibility. Although they weren't the first to use GPUs to train CNNs, their beating ImageNet challenge by a large margin was a hallmark event that made people focus attention to deep learning - ThisFeelsABitOff (talk) 20:47, 28 October 2021 (UTC)

I have added Alexnet, and also word2vec. I may add ELMO/Bert, what do you think? WillSmith (talk) 21:04, 12 September 2023 (UTC)

References

New achievements?

I'm starting a new topic because the OP of the other one didn't sign it, there's no reply button and editing is clunky.

In the last few years we've had lots of "milestones" claimed by ChatGPT/OpenAI, ClaudeAI/Anthropic and the like. Are they worthy of inclusion? TurboSuperA+ (talk) 16:59, 28 December 2024 (UTC)

Does "super-Turing computation" belong in this table?

This is the first time I have heard of this term, and from my understanding it had no substantial impact on machine learning. I'm surprised it is listed here in the 90s. Morinator (talk) 23:56, 13 March 2025 (UTC)

Proposed new timeline entries

with all of the releases and advancements, it’s a bit difficult to choose exactly which models and developments are worth inclusion.

list below are my candidates for inclusion.

Year	Type	Entry
2024	Achievement	OpenAI o1 reasoning model — OpenAI releases the o1 series, beginning with o1-preview on September 12 and the full model on December 5. The models introduce large-scale inference-time computation ("chain-of-thought" reasoning), spending additional compute on each query before responding. Performance on math and coding benchmarks improved substantially over GPT-4o. OpenAI described o1 as the first in a planned series of reasoning models.
2024	Achievement	Nobel Prize for AlphaFold -The Nobel Prize in Chemistry was awarded to Demis Hassabis and John Jumper (Google DeepMind) for protein structure prediction with AlphaFold, and David Baker for computational protein design. It was the first Nobel Prize given for work in machine learning.
2024	Release	DeepMind Genie 2 — Google DeepMind announces Genie 2 on December 4, a foundation world model that generates interactive 3D environments from a single image prompt. Trained on video data, the model can simulate object interactions, physics, character animation, and NPC behavior. DeepMind positions it as a research tool for training and evaluating embodied AI agents.
2025	Release	DeepSeek R1 — Chinese AI company DeepSeek releases R1 on January 20, an open-source reasoning model with 671 billion total parameters (37 billion active per forward pass) built on a mixture-of-experts architecture. R1 achieved performance comparable to OpenAI's o1 on math, coding, and reasoning benchmarks.
2025	Release	Anthropic Claude 4 — Anthropic releases Claude Opus 4 and Claude Sonnet 4 on May 22, built around hybrid reasoning: near-instant responses for simple queries, and extended thinking for harder problems. Both models designed to support tool use during extended thinking. Anthropic deploys Opus 4 under its ASL-3 safety standard, a first for the company.
2025	Release	Google Gemini 2.5 series — Google releases Gemini 2.5, starting with 2.5 Pro in March and 2.5 Flash in April, with general availability following in June. Both are "thinking models" with controllable reasoning budgets that let developers set how many tokens a model spends deliberating before it responds. 2.5 Pro includes a "Deep Think" mode for complex math and coding. The 2.5 family supports native multimodality, and code execution.
2025	Release	Meta Llama 4 — Meta releases Llama 4 on April 5, its first model family to use a mixture-of-experts architecture and its first with native multimodality. The release includes Scout (17B active parameters, 16 experts) and Maverick (17B active, 128 experts, 400B total). Scout offered a 10-million-token context window. Behemoth, a nearly 2-trillion-parameter teacher model, is previewed but not publicly released.
2025	Release	OpenAI GPT-5 — OpenAI releases GPT-5 on August 7, a unified model that routes queries between a fast response mode and a deeper reasoning mode.
2025	Protocol	Agent2Agent (A2A) and Model Context Protocol (MCP) — Anthropic open-sources the Model Context Protocol (MCP) in November 2024, a standard for connecting AI models to external tools and data sources via JSON-RPC. OpenAI adopts MCP in March 2025, Google DeepMind in April. Google then launches the Agent2Agent (A2A) protocol on April 9, 2025. A2A handles agent-to-agent collaboration (discovery, task delegation, and coordination), complementing MCP's model-to-tool connectivity.

StandardDeviatedSeptuagenarian (talk) 01:40, 3 April 2026 (UTC)