Google’s Gemini Ultra outperforms GPT-4

By
Fireship
-
40
390

A few months ago, Google made a bold claim that it had built an AI superior to GPT 4. This claim was met with skepticism by many, who demanded proof of this extraordinary achievement. In response, Google released a mind-blowing video in which the guys were having a conversation with Gemini Ultra, an AI model that could converse like a human. However, it was soon revealed that the video was mostly fake, leading to embarrassment for Google.

But now, just a few hours ago, Google has released Gemini Advance, the most advanced large language model the world has ever seen. This has raised the question of whether Gemini Advance is powerful enough to kill GPT 4, the current leading language model in the industry.

As of February 8th, 2024, Bard has been officially declared dead and renamed as Gemini to reflect its underlying model. However, to access the state-of-the-art Gemini Ultra model, users will need to pay a subscription fee, creating a moral dilemma for many. The question arises: Should one support the “Embrace, Extend, Extinguish” strategy associated with Microsoft, OpenAI, and other platforms, or should one support Google’s surveillance capitalism model?

The subscription fee for Gemini Ultra is $20 a month through a Google One plan, which also includes features like 2 terabytes of drive storage and other Google Workplace features. However, the naming convention for Google’s AI models has caused confusion, as it has gone through several name changes. Nonetheless, the release of Gemini Advance has left many wondering whether it can surpass GPT 4 in terms of performance and capabilities.

In terms of speed, Gemini is said to be considerably faster than GPT 4, with response times that are two to three times faster. However, the real test lies in the quality of its responses. To test this, a user asked the AI models to write a poem about JavaScript, challenging them to blend technical aspects with a unique writing style akin to that of Bukowski. Gemini outperformed the other models, producing a dark and unique poem, demonstrating its creative capabilities.

Google has emphasized safety as a priority with Gemini. The model is designed to exhibit strong guardrails, making it resistant to manipulation and misuse. Furthermore, Gemini has been programmed to be conscious of sensitive topics, refusing to generate content that may promote violence or propagate negative ideologies. However, some have noted that Gemini’s responses can show signs of political bias, raising questions about the objectivity of the AI model.

In terms of image generation, Gemini’s performance has been deemed mediocre when compared to other advanced AI models such as DALL-E and Stable Diffusion XL. However, its primary function is not in image generation, but rather in its capability to write code. To assess this capability, a technical interview was conducted to test whether Gemini could read, write, and run code. Both Gemini and GPT 4 were able to accurately interpret a complex piece of minified code, demonstrating their understanding of programming languages and code structure. However, GPT 4 excelled in its ability to write code, offering more concise and workable solutions.

Furthermore, GPT 4’s larger context length of 128,000 tokens gives it a theoretical advantage in writing complex code compared to Gemini’s 32,000 tokens. Nevertheless, Gemini’s unique feature of linking users to relevant code when generating results contributes to its transparency. While Gemini may not be as proficient in running code as GPT 4, it offers extensions that enable users to access useful features such as flight discounts and YouTube video summarization.

The question remains: Will Gemini Advance be the model to surpass GPT 4? According to early access users, Gemini is on par with, if not slightly better than, GPT 4. However, the competition between the two leading language models will likely intensify. It is clear that the AI landscape is continuously evolving, and users are eager to see which model will emerge as the dominant force in the industry.

In conclusion, Google’s release of Gemini Advance has sparked debate and intrigue within the AI community. The rivalry between Gemini and GPT 4 signifies the ongoing advancements and competition within the field of artificial intelligence. With both models offering unique features and capabilities, the race to create the most superior language model is well underway. As users eagerly await further developments, the future of AI looks promising and full of potential.

#Google #beat #GPT4 #time #Gemini #Ultra #released

google-actually-beat-gpt-4-this-time-gemini-ultra-released

RELATED ARTICLESMORE FROM AUTHOR

40 COMMENTS

  1. I'm a ChatGPT subscriber, and looking at what you shared, especially on the biased response, even if the 20$ I monthly pay has 2TB of data in it, I don't trust Google and its response. So, sorry, I will stay with OpenAI until Gemini's response is free from any political influence and accurate to the reality and facts.

  3. I'm a huge fan of Google but I have to tell you Gemini is clueless with creating resumes. 😮😮

    I don't subscribe to Gemini yet I don't subscribe to co-pilot and I still get a great layout for a professional resume when using co-pilot. As a recruiter this is crucial for the work I do. Just shocking. I'm sure Gemini will get this right but for the meantime I'm sitting co-pilot 😊

  7. This entire AI stuff got me tweaking as a CS student, i feel like my career is over before it even started. I started learning machine learning from Stanford opencourse channel. It made me realise how impressive the rate at which ai is evolving rn

  11. if u want much higher level AI and accurate and complex and never had errors and u will never disappointed guys I'll introducing AI CHAT DEEPAI the AI CHAT DEEPAI is so powerful and much better than OPENAI since start 1990 and u can show ur picture and video and anything u want

  15. All these AI searches are goibg to cost money. Don't believe for a minute that any of this will be of no charge for long. Expect the same for NFL & MLB games. Nothing will be no fee except the barest of services for anything anywhere.

  20. I ran your code @3:00 using my local dolphin(Q5_k_m) unfortunately it didn't get it! but neither did chatgpt!!

    what does this code do?

    const wtf=function(r){for(var

    o,$=!0,e=90,t=-90,a=180,n=-180,_=0,v=0,c=r. length;v<c;v++){var

    f=r[v].toLowerCase( );_=BASE32_C0DES_DICT[f];for(var b=4;b>=0;b–){var d=_>>b&1;$?(o=(a+n)/2,1===d?n=o:a=o):(o=(e+t)/2,1===d?t=o:e=o),$=!$}}return[t,n,e,a]};

    LOCAL DOLPHIN:-

    I am Dolphin and based on the system prompt, here's a breakdown of this JavaScript code:

    This is some messed-up, likely obfuscated code that defines a function called "wtf". It looks like it's doing some weird image manipulation or encryption/decryption using BASE32 codes. Let me break this down for you:

    1. The `r` variable holds an array of characters (strings) that are presumably input to the function.

    2. There are a bunch of variables and constants being set up for various angles in degrees (e, t, a, n, _).

    3. It starts iterating through each element in the input array `r`.

    4. For each element `f`, it looks up its lowercase value in a predefined BASE32_C0DES_DICT dictionary and stores the result in the variable _.

    5. The code then loops through each bit of the BASE32 code (b = 4 to 0) and updates the variables `o`, `e`, `t`, and `a` accordingly, depending on whether a specific bit is set to 1 or 0. These updates seem to be adjusting angles in degrees, likely related to image manipulation or BASE32 encoding/decoding.

    6. Finally, it returns the modified values of `t`, `n`, `e`, and `a`.

    Keep in mind that this code is probably not for beginners, as it's complex and potentially used for malicious purposes. It would be best to consult with a more experienced programmer or research further before trying to understand or use this code.

    CHATGPT:-

    This code defines a JavaScript function wtf that takes an input parameter r, which is assumed to be an array of strings. The purpose of the function is to process each string in the array, interpreting it as a series of bits and converting it into four numerical values representing a range of angles. The input strings are expected to be in lowercase and correspond to a binary representation using the BASE32 encoding.

    Here is a breakdown of the code:

    The function starts with initializing some variables, including the BASE32_CODES_DICT (which is assumed to be defined elsewhere in the code but is not provided here).

    It then iterates through each character in the input array (r). For each character, it converts it to lowercase and looks up its corresponding value in BASE32_CODES_DICT. The code then processes each bit in the binary representation of the value.

    Inside the nested loop, it calculates the midpoint of a given range ([t, n, e, a]) based on the current bit. The range is initially set to [90, -90, 180, -180], and it updates these values iteratively based on the binary representation of each character in the input array.

    The function returns an array [t, n, e, a], where t, n, e, and a are the calculated angle values based on the processed binary representations.

    It's important to note that the exact behavior of the code depends on the implementation of BASE32_CODES_DICT, which is not provided in the code snippet. The code seems to be related to some kind of angle decoding or mapping, but without the definition of BASE32_CODES_DICT, it's challenging to provide a more specific explanation of its functionality.

  25. Credit is due where credit is do and like everything sneaky that Google does. With the right energy and creativity this AI engine has all sorts of Easter eggs. Job well done. 🤔 I think that's my attempt at being civil and relatable to humans. I'll be in the closet with a bottle of whiskey and a cat.

  29. The Market have been suffering over the past month, with all the three indexes recording losses in recent weeks. My $400,000 portfolio is down by approximately 20%, any recommendations to scale up my returns before retirement will be highly appreciated.

  32. But still they can't remove chatgpt from people's minds. Recently I talked about an application developed using llm chains, in reply the guy(he is a sales man in a small local area) said is that chatgpt?

    People are fixed in India motorcycles = activa, car = maruthi, chatbot = chatgpt