Just now, Google made a big move late at night - Bard is now collectively known as Gemini. releaseGeminiAdvanced, supported by Google's most powerful multi-modal large model GeminiUltra1.0! In order to prevent conceptual confusion, we use OpenAI’s large model to compare and understand:
Gemini is the general name of the brand, equivalent to OpenAI’s ChatGPT;
GeminiAdvanced paid service, corresponding to ChatGPTPlus;
GeminiUltra model, the benchmark is GPT-4;
The operation of switching to the GeminiAdvanced interface is similar to ChatGPT. Just click on the option in the upper left corner:
In terms of price, GeminiAdvanced’s pricing is also quite interesting——$19.99/month, cheaper than ChatGPTPlus ($20/month).
However, Google also gives a small benefit. In the first two months after subscribing,waived! fee! !
And there is no limit on the number of uses per hour like GPT-4, so you can use it freely.
Not only that, Google also incidentallyAndroidLaunched on mobile phoneGemini APP, the kind that specific Android users can "directly access" by pressing and holding buttons such as the power button:
iOSUsers don’t need to worry, Gemini will appear in Google App in the next few weeks, and the opening method will be as follows:
After all, when Google previously released GeminiUltra, a large model, it scored 30 SOTA in 32 benchmark tests and was the first to reach the level of human experts on the MMLU benchmark.
Now that it has been commercialized and finally launched, many netizens flocked to it, and there was also a brief downtime.
So what is the effect of GeminiAdvanced, which is powered by Google's most powerful model?
We successfully launched the trial in the first time.
Measured GeminiAdvanced
Although Google states that it currently only supports English, actual testingAsk questions in Chinese and it will not only understand but also answer in Chinese.
Since it is produced by Google, it must test its Internet search capabilities.
I originally wanted to see if it could be used as a melon-eating tool, but due to Google's strict ethical and moral restrictions, GeminiAdvanced refused to answer on the spot.
Then the next best thing is to ask pure facts without value judgments, and its performance will be very impressive.
Answering,Expand the statements marked in green to see the source of the citation..
Statements marked in yellow indicate that no clear source of citation was found., you can try to verify further.
For table data generated by AI in the answer, you can also click "Export to Sheets"One-click import to GoogleDocs for further editing and processing, it can be said to be very practical in work scenarios.
Next, you can also turn on support for other Google services in "Extensions", such as maps, Gmail, and YouTube videos.
After you associate your email account, GeminiAdvanced becomes your personal AI butler, which can help you manage many things, such as finding and unsubscribing from spam emails.
Unfortunately, the extension does not currently support Chinese commands.
Using "findmeyoutubevideos..." in English can trigger the search video function, which is also a good tool for assisting in learning knowledge.
In addition to Internet search and integrated applications, Google also particularly emphasizes GeminiAdvanced's reasoning capabilities.
Let's start with a classic reasoning question from Microsoft's test of GPT-4. As a result, GeminiAdvanced not only successfully answered it, but also considered additional low-probability situations.
Pay attention to the interface"Showdrafts"Button, representing GeminiAdvancedThree "drafts" are generated each time, and select the best of them to display.
The three drafts used different reasoning methods or experimented with different writing styles, but the answers were invariably correct.
If by chance you are not satisfied with all three drafts, you can also select the Resume All button on the far right.
Pay attention to the last row of buttons in the answer. In addition to the regular likes, dislikes, and shares, there are alsoTwo uncommon new features.
The slider button in the middle represents "Edit Answer", you can choose shorter, longer, simpler explanation, lighter tone, or more formal tone.
Try choosing a lighter tone and your overall answer will become more colloquial.
If you choose to be more formal, the entire answer will be like answering a paper in an examination room.
finalThe GoogleG icon represents the use of search engines to verify whether answers generated by AI are accurate., the results will also be marked in the form of "green - with cited sources" and "yellow - without cited sources".
In a more practical scenario, if it is required to generate a technology-themed Spring Festival couplets, GeminiAdvanced can also meet the requirement of "straight up and flat".
Generating the code is a piece of cake, and it also hides its own advertisements.
After some experience, here is a final summary.
GeminiAdvanced is supported by the extra-large GeminiUltra model.Ability basically reaches the same level as GPT-4.
The design is more like a mature product rather than a large-model technology demonstration demo.
After integrating with Google's powerful Internet services, it is also unique in practicality.
In addition, before this release, Qubit also had a brief exchange with the Google Gemini team.
The team said that this release is more focused on releasing GeminiUltra’s language capabilities into the product.In the future, we will continue to update multi-modal capabilities, more interactive code functions, and upload file analysis data and other functions..
Deeper integration with Google products, such as using Gemini directly in Gmail to reply to emails, which is currently in the "Comingsoon" state.
But we are going to get a schematic, so stay tuned.
In addition, during the exchange, the Google development team reminded one thing:
Since the product has just been renamed from Bard to Gemini, AI will occasionally be confused and it will take time to transition slowly.
It turns out that AI will not adapt to changing its name just like humans, which is also hilarious.
Both are $20, which one do you pick?
Just when the news about GeminiUltra came out, the well-known breaking account Flowersfromthefuture organized a vote.
For the same $20, which one would you subscribe to, GPT-4 or GeminiUltra?
In the end, 2,360 people participated, 40% of whom firmly decided to stay in GPT-4, and only 12.3% chose to migrate to GeminiUltra.
But this vote comes days before the actual release.
After experiencing it, I don’t know how many people will “abandon O and choose G” because of the product’s functional experience and service integration.
Just now, a professor at Wharton Business School said that he had experienced GeminiAdvanced in advance for 6 weeks.
One comment he gave was:
GeminiAdvanced is clearly at the level of GPT-4, but does not significantly exceed it.
Both have their own advantages during use.
For example inSearch capabilitiesFor one thing, let's both check out the latest sneaker trends. GeminiAdvanced searches YouTube, while ChatGPT uses Bing.
This shows that GeminiAdvanced is different in terms of search integration.
The professor also believes that GeminiAdvanced’s interface is smoother than GPT-4, and there are fewer technical errors.
It differs in "personality" from GPT-4, being more friendly and willing to engage in word play. Despite their personality differences, the two showed compatibility in processing complex cues.
Of course, this is not a direct comparison between GeminiAdvanced and ChatGPT, but a discussion of the possible future development directions of AI through the two:
The unique advantages and disadvantages of GeminiAdvanced compared to GPT-4 indicate that there is still a lot of room for improvement in the model, and we will continue to see rapid progress in the future. The wave of AI development has not yet reached its peak, and OpenAI’s next step may be to release the rumored GPT-4.5 or GPT-5.
Now more than 14 months have passed since the release of ChatGPT, and Google has finally prepared competing products in terms of models, applications, and ecology.
However, the old rival OpenAI has quietly rushed to the next battlefield.
According to The Information,OpenAI is developing a new generation of Agent applications, move the cursor, click, enter text and use various APPs like humans according to user requests.
Such as filling in data from documents into spreadsheets for analysis, or automatically filling in expense reports in accounting software.
In other words, the next generation of ChatGPT will take over your phone and computer.
Reference links:
[1]https://blog.google/technology/ai/google-gemini-update-sundar-pichai-2024/
[2]https://blog.google/products/gemini/bard-gemini-advanced-app/
[3]https://www.oneusefulthing.org/p/google-gemini-advanced-tasting-notes
[4]https://www.theinformation.com/articles/openai-shifts-ai-battleground-to-software-that-operates-devices-automates-tasks