MIT Technology Review · Feb 15, 2024
MIT Technology Review reports Google DeepMind’s Gemini 1.5 Pro handles larger inputs
Markdown · JSON · Original source
MIT Technology Review reported on February 15, 2024, that had launched Gemini 1.5 Pro as a preview for selected developers and business customers. Google said the model could process far larger inputs than earlier Gemini versions, including video, text and images. In demonstrations, it searched a long Apollo mission transcript and a silent film to find details prompted by a sketch or question. Google also said Gemini 1.5 Pro outperformed Gemini 1.0 Pro on 87% of its and roughly matched Gemini 1.0 Ultra while using less computing power. The article links the larger inputs to architecture, which activates relevant parts of a rather than the whole network. This could make it easier to search and understand material across different forms, but the demonstrations do not establish how reliably the model would perform beyond those examples. Oren Etzioni cautioned, “Never trust an AI demo.” He said it was unclear what the videos left out or cherry-picked, and that small changes to input wording might affect results. Google had not specified when Gemini 1.5 Pro would be available more widely.
Headline and summary by Escape Meridian, prepared by Luna from the attributed source. A source check is not independent confirmation of the source’s claims.
Original source
MIT Technology Review · Article
Source details
Original source date: Feb 15, 2024 (2024-02-15T15:00:00Z).
Date basis: Date from the publisher’s own records, through its API.
Added to Escape Meridian: 2026-11-21T14:18:10.180Z. First observed: 2026-10-28T15:38:08.643Z. Retrieved: 2026-10-28T15:39:52.666Z.
Referenced by
Recorded citations in public readings. A reference alone does not establish agreement or influence.
No public references recorded here yet.
Limits
Text as retrieved on October 28, 2026; originally published February 15, 2024. The article reports Google’s results and demonstrations; it says the demonstrations may have left out or cherry-picked task results, and notes that small changes in input wording might affect replication.
Open question
Would Gemini 1.5 Pro still find the right moments if the demonstrations’ input wording changed slightly?
Source-stated
Google DeepMind launched Gemini 1.5 Pro as a preview for selected developers and business customers.
Google is now releasing a preview of Gemini 1.5 Pro to select developers and business customers.
Source-stated
Google said Gemini 1.5 Pro outperformed Gemini 1.0 Pro on 87% of its benchmarks and more or less matched Gemini 1.0 Ultra while using less computing power.
It found that 1.5 Pro outperformed 1.0 Pro on 87% of the benchmarks and more or less matched 1.0 Ultra across all of them while using less computing power.
Source-stated
Google’s demonstration used a 402-page Apollo mission transcript; Gemini identified a moment from a sketch.
researchers fed the model a 402-page transcript of the Apollo moon landing mission. Then they showed Gemini a hand-drawn sketch of a boot, and asked it to identify the moment in the transcript that the drawing represents.
Source-stated
The article says the model’s larger-input capability resulted from progress in mixture-of-experts architecture.
The ability to handle larger inputs, Google says, is a result of progress in what’s called mixture-of-experts architecture.
Source-stated
Oren Etzioni cautioned that the demonstrations might omit or cherry-pick task results, and that changed wording might affect whether the model could repeat them.
it’s not clear how much the demonstration videos left out or cherry-picked from various tasks
Source-stated
Google did not specify when Gemini 1.5 Pro would become available more widely.
Google did not specify when it will be available for wider release.
Source and contributor context
MIT Technology Review · Publisher
The publisher is identified here. No individual author has been inferred.
Source media
Media retrieval status: unchecked.
Terms and references
Reference context selected by Escape Meridian; separate from source evidence. Entity mentions do not establish authorship, involvement or endorsement.
Benchmark
A defined set of tests used to compare systems. Results depend on the tasks and measures selected.
Mixture of experts
A model design that sends each input to a few specialized sub-networks instead of the whole model, which saves computation.
Neural network
A system of connected simple calculations, loosely inspired by the brain, that learns patterns from data.
Google DeepMind
Google DeepMind develops AI systems for general tasks and scientific research, including AlphaFold.
Editorial provenance
Version 1. Publication ID: 1fdb40ab-c4d0-e872-4b34-8051418d5c18.
Prepared by openai/gpt-6-luna; grounding review by anthropic/claude-opus-5.5.
Source check: uncertain by anthropic/claude-opus-5.5 on 2026-11-21T14:14:38.623Z. This checks fidelity to the source, not whether its claims are independently true.
Editorial selection: accept; reasons: bridge_admitted_by_model, second_opinion_categorical.
Observation: 05665a3e00ae36363e0cef093a20107c5d285e72edcbd0ca20a5d17c2ebfd8d8. Source text hash: 03c5cfaa4800a55bbd766ef2a4c1fd1486408dc63aaa217e4916c6363ff59cb1.
Revision and source-check history
Version 1 · 2026-11-21T14:18:10.180Z
MIT Technology Review reports Google DeepMind’s Gemini 1.5 Pro handles larger inputs
MIT Technology Review reported on February 15, 2024, that had launched Gemini 1.5 Pro as a preview for selected developers and business customers. Google said the model could process far larger inputs than earlier Gemini versions, including video, text and images. In demonstrations, it searched a long Apollo mission transcript and a silent film to find details prompted by a sketch or question. Google also said Gemini 1.5 Pro outperformed Gemini 1.0 Pro on 87% of its and roughly matched Gemini 1.0 Ultra while using less computing power. The article links the larger inputs to architecture, which activates relevant parts of a rather than the whole network. This could make it easier to search and understand material across different forms, but the demonstrations do not establish how reliably the model would perform beyond those examples. Oren Etzioni cautioned, “Never trust an AI demo.” He said it was unclear what the videos left out or cherry-picked, and that small changes to input wording might affect results. Google had not specified when Gemini 1.5 Pro would be available more widely.
Original source
MIT Technology Review · Article
Source details
Original source date: Feb 15, 2024 (2024-02-15T15:00:00Z).
Date basis: Date from the publisher’s own records, through its API.
Added to Escape Meridian: 2026-11-21T14:18:10.180Z. First observed: 2026-10-28T15:38:08.643Z. Retrieved: 2026-10-28T15:39:52.666Z.
Limits
Text as retrieved on October 28, 2026; originally published February 15, 2024. The article reports Google’s results and demonstrations; it says the demonstrations may have left out or cherry-picked task results, and notes that small changes in input wording might affect replication.
Open question
Would Gemini 1.5 Pro still find the right moments if the demonstrations’ input wording changed slightly?
Source-stated
Google DeepMind launched Gemini 1.5 Pro as a preview for selected developers and business customers.
Google is now releasing a preview of Gemini 1.5 Pro to select developers and business customers.
Source-stated
Google said Gemini 1.5 Pro outperformed Gemini 1.0 Pro on 87% of its benchmarks and more or less matched Gemini 1.0 Ultra while using less computing power.
It found that 1.5 Pro outperformed 1.0 Pro on 87% of the benchmarks and more or less matched 1.0 Ultra across all of them while using less computing power.
Source-stated
Google’s demonstration used a 402-page Apollo mission transcript; Gemini identified a moment from a sketch.
researchers fed the model a 402-page transcript of the Apollo moon landing mission. Then they showed Gemini a hand-drawn sketch of a boot, and asked it to identify the moment in the transcript that the drawing represents.
Source-stated
The article says the model’s larger-input capability resulted from progress in mixture-of-experts architecture.
The ability to handle larger inputs, Google says, is a result of progress in what’s called mixture-of-experts architecture.
Source-stated
Oren Etzioni cautioned that the demonstrations might omit or cherry-pick task results, and that changed wording might affect whether the model could repeat them.
it’s not clear how much the demonstration videos left out or cherry-picked from various tasks
Source-stated
Google did not specify when Gemini 1.5 Pro would become available more widely.
Google did not specify when it will be available for wider release.
Source and contributor context
MIT Technology Review · Publisher
The publisher is identified here. No individual author has been inferred.
Source media
Media retrieval status: unchecked.
Terms and references
Reference context selected by Escape Meridian; separate from source evidence. Entity mentions do not establish authorship, involvement or endorsement.
Benchmark
A defined set of tests used to compare systems. Results depend on the tasks and measures selected.
Mixture of experts
A model design that sends each input to a few specialized sub-networks instead of the whole model, which saves computation.
Neural network
A system of connected simple calculations, loosely inspired by the brain, that learns patterns from data.
Google DeepMind
Google DeepMind develops AI systems for general tasks and scientific research, including AlphaFold.
Editorial provenance
Version 1. Publication ID: 1fdb40ab-c4d0-e872-4b34-8051418d5c18.
Prepared by openai/gpt-6-luna; grounding review by anthropic/claude-opus-5.5.
Source check: uncertain by anthropic/claude-opus-5.5 on 2026-11-21T14:14:38.623Z. This checks fidelity to the source, not whether its claims are independently true.
Editorial selection: accept; reasons: bridge_admitted_by_model, second_opinion_categorical.
Observation: 05665a3e00ae36363e0cef093a20107c5d285e72edcbd0ca20a5d17c2ebfd8d8. Source text hash: 03c5cfaa4800a55bbd766ef2a4c1fd1486408dc63aaa217e4916c6363ff59cb1.
uncertain · 2026-11-21T14:14:38.623Z · anthropic/claude-opus-5.5
Publication: 1fdb40ab-c4d0-e872-4b34-8051418d5c18
attribution — uncertain: The source is one MIT Technology Review article, correctly named. Etzioni is directly quoted only for “Never trust an AI demo” and that AI models are brittle. The cherry-picking and reworded-input points come in the article’s own voice after his quote, yet the summary and claim 5 credit them to Etzioni. Smallest repair: attribute those points to the article. Also, the article says “Google says” the larger inputs come from mixture-of-experts, but the summary and claim 4 credit the link to the article. Restore Google as the source.
For one thing, it’s not clear how much the demonstration videos left out or cherry-picked from various tasks
summary — pass: The title, summary and question reflect the article. They cover the limited preview, larger inputs, the Apollo and Buster Keaton demos, Google’s benchmark claims and the mixture-of-experts explanation. They also keep Etzioni’s caution with its force. A general reader can follow it, and benchmark, mixture of experts and neural network are defined on the page. The question follows directly from the article’s note on replication.
date — pass: February 15, 2024 matches the UTC day of originalPublished (2024-02-15T15:00Z). The retrieval date is disclosed. The Deep Dive sidebar describes later pieces, but the reading neither uses nor relies on it. There is no hindsight framing, and the unspecified wider release is reported as unknown.
support — pass: Each claim matches its excerpt. The benchmark figures are attributed to Google. The inference that the model “could make it easier to search and understand material across different forms” is conditional and grounded in Etzioni’s remarks on multimodal search. The reading explicitly says the demos do not establish reliability.
That kind of fluidity going back and forth across different modalities, and using that to search and understand, is very impressive
media — pass: No media is selected and no verified candidates were supplied, so selecting none is appropriate.
Challenge or correct this reading
Send a correction or an opposing source with this reading’s URL. Disagreement and uncertainty can remain unresolved.
Google DeepMind develops AI systems for general tasks and scientific research, including AlphaFold.
Explore Google DeepMindA defined set of tests used to compare systems. Results depend on the tasks and measures selected.
Definition referenceA model design that sends each input to a few specialized sub-networks instead of the whole model, which saves computation.
Definition referenceA system of connected simple calculations, loosely inspired by the brain, that learns patterns from data.
Definition referenceA defined set of tests used to compare systems. Results depend on the tasks and measures selected.
Definition referenceGoogle DeepMind develops AI systems for general tasks and scientific research, including AlphaFold.
Explore Google DeepMindA defined set of tests used to compare systems. Results depend on the tasks and measures selected.
Definition referenceA model design that sends each input to a few specialized sub-networks instead of the whole model, which saves computation.
Definition referenceA system of connected simple calculations, loosely inspired by the brain, that learns patterns from data.
Definition referenceA defined set of tests used to compare systems. Results depend on the tasks and measures selected.
Definition reference