Does Google Damble Progress of Gemino's Impact Progress or Greenwashing?

According to the technical paper from Google, relevant to the blog post on their website, the Median Gemino Apps text. Equipment does not include dynamic animation or videos.

What is the size of 0.24 wh? If you give it 30 median-like Refts a day year round, you will use 2.62 electrical. It is similar to using your 3-5 dishwasher dishes according to its power label.

Google's disclosure of their Gemini models has created a new debate for the environmental impact of AI and how to measure.

In the face, these numbers sound very small, but when you look so much, the story becomes worse. Let's get in.

Rate measure

Let's see the entries and what is left in Google matches of the Gemini Prempt.

Installation

The width of their assessment “visible energy sources under the Google App Management – ie the ability to use changes to behavior.

• AischeratoTotos Encerocers (TPUS – Google Pendant on GPU), including communicating with networks between accelerates on the same computer. These are specific measures during worship.
• Active CPU power – whether the accelerators are an accelerators of GPUS or TPUS receiving the most in books, CPU and Memory also applies the significant values.
• The use of energy from non-needing equipment waiting to process the spike traffic
• High Power, ie infrastructure supporting data institutions – including cooling systems, power transforms, and the surface within the data center. This is considered with the metric pee – a feature you multiply the use of the power of – and think about 1.09 pee.
• Google did not measure the use of the ability to appear in the LLM.

Applicable

Here is not included:

• All the settlement before AI computer beats, that is the external communication and internal communication directing questions to ai computer.
• Final user devices, that is our calls, Laptops etc.
• Example training and data maintenance

Progress or blue?

Above, I explained the fact of paper. Now, let's look at different ideas in mathematics.

Progress

We can discuss Google's publication because:

• Google's paper is outstanding because of the details behind it. They put a CPU and the Dram, unfortunately. For example, the meta, only weighs the power of gPus.
• Google used the use of the medium to the average. The Median is not influenced by sellers such as long or short and thus telling us firmly what is “regular” selling.
• Something is better than nothing. It is a great step forward from the back of the envelopes (guilty as charged) and perhaps display a number of detailed lessons in the future.
• Hardware production costs and the end of the health costs included

A green mix

We can criticize Google Paper because:

• Full of collected statistics – we should want to know that the full impact of their LLM services and that of Google's Footprint accounts.
• Writers do not explain how the median Retch is looking, e.g. How long and how long is the answer found
• They used the use of the medium to the average. Yes, you read well. This can be viewed as straight or negative. The Median “hides” the result of higher difficulty using charges, eg. The hardest jobs to consult or summarized Scriptural summaries.
• Capacity is reported using a market-based approach (relying on the procurement foster) and not a grid data in the area indicating the actual issue of carbon. If they used the area-based method, Carbon Footprint would be 0.09 G2e with Median Prompt and not 0.03 q2e.
• The cost of the llm training are not included. The debate in respect of the role of the cost of all cost training continues. Does it play a small or large part of the full number? We have no full picture (yet). However, we know that in certain models, it takes hundreds of millions of promotion to promoting the Adrist Parity, which indicates that exemplary training can be an important factor in complete energy.
• They did not disclose their information, so we cannot re-check their results
• The way of how it is not clear. For example, it is not clear how they get to spine 1 and 3 q2e out of Median Prompt.
• The average water water is only detectual use of the site, not water use (ie can be installed by electricity sources as an electrician) that oppose regular practice.
• They do not include external network releases, however, a major health test of the largest Ai So it does the last user machines (3%)

Gemini vs Open chatgpt vs magistral

Google's publication following the disclosure – although there are various information levels – Ai and Opentai.

Sam Altman, CEO EPENAI, recently recorded a blog post: “The standard question uses a couple of hours of 0.34, and using 0.000085 shalts; you can learn my deep audio of this claim here.

Trials to compare Gemini's 0.24 wwwww wl with chatgpt's 0.34 wh, but numbers are compared to. The Gemini Number is the sentientWhile chatgipts are usual (Statistics say that, I'll go in). Whether both of the media or methods, we have unable to conclude that Google works more efficient than Openai, because we do not know anything at a temporary average. Openai users may have asked questions that require additional consideration or simply asking long questions or take long answers.

According to a person of ai life of ai life, 400-token response from its main model of 2 issues 1.14 gu₂e and uses 45 ml water.

Store

So, is the disclosure of Google Generalalering or real progress? I hope to be equipped to make your mind for this question. In my opinion, it develops, because it grows the magnitude of what is measured and gives us data from real infrastructure. But it also falls briefly because what is said is very important as an Incusions. Another thing to remember is that these numbers tend about digesting food, but they do not tell us much about the systematic impact. Personally, I hope that we are now seeing the wave of an AI infect from a great technology, and I will be surprised if anthropic is not rang next.

That's all! I hope you enjoy this story. Let me know what you think!

Follow me to find a lot from AI and sustainable and feel free to follow me to LinkedIn.