AI Will Change the World in 2024
The year 2024 is arrived, and I have a ton of amazing plans for us. Think that 2023 was a significant year for AI? Wait until you see
what 2024 has in store. In light of the fact that AI will undergo a sea change this year, these are my ten predictions for AI in 2024.
1 – LLaMA 3
LLaMA 3 will be released in the first half of 2024, according to my original guess. The main feature of LLaMA 3’s release is its ability to close the gap between cutting-edge proprietary models like GPT4 and open-source models.
Not even a full year has passed since the release—or rather, leak—of LLaMA 1, and when we combine that with Meta’s publication of LLaMA 2, we have a huge impact on the future of open-source models in the AI field.
While Mark Zuckerberg has already hinted at the release of LLaMA 3 in 2024, Meta’s current focus is on improving LLaMA 2 for integration into their consumer products,
which is extremely impressive in and of itself. Meta’s dedication to open-source has established the company as a major player in the AI industry. If you were under the impression that LLaMA 2 was not nearly as powerful as GPT 4,
Meta’s integration of it into their products—which cater to billions of users—proves that it is production-ready.
To further build anticipation for Meta’s contributions to open-source AI in 2024, Zuckerberg has hinted at AI Studio, a platform that lets users construct AI models just as readily as user-generated content on Facebook.
2-Gemini Ultra,
My second prediction, which is obviously untrue, is that Google will unveil Gemini Ultra in 2024. By the end of 2023, Gemini gained popularity
thanks to a number of research publications and demo films demonstrating its equal competence as the GPT 4 model. However,
Gemini Ultra caused controversy when it was revealed through a leak that the video was heavily altered and did not accurately showcase Gemini’s abilities. Still,
the research paper’s descriptions of the performances are quite accurate, so I’m very interested to see what Gemini Ultra has to offer. I want further competition in the closed-source, proprietary model market,
and Google will present some formidable opposition to GPT4.
Gemini is available in three packages: Ultra, Pro, and Nano. Nano will be entirely contained within the device and operated without the need for an internet connection.
Gemini Pro will function more like an LLaMA 2 and require substantial consumer gear. Only cloud servers will be able to operate Gemini Ultra.
In order to make sure developers are using the Gemini models, Google intends to make a significant investment. OpenAI, too, wants developers to build upon Gemini with GPT4.
That’s how you create a vibrant developer ecosystem: by providing an extremely attractive developer product.
Since Apple already has a robust developer community, Google Gemini and Open AI GPT4 will have to compete harder to win over developers.
The victor amongst Apple, Meta, Google, and Microsoft OpenAI will be the one who can entice the most number of developers to create amazing apps on top of their AI platforms.
I think that Google Gemini Ultra will be of the highest caliber. I believe it will have some initial issues, including vulnerabilities, hallucinations, and bugs, but once users and developers have access to it, it will fast grow better.
3 – Robots
The next thing I predict will happen is that robots will advance greatly. In 2024, more businesses will provide robot products, and humanoid robots and
other forms of robots will keep developing and getting better. While Boston Dynamics is still a major contender, Tesla has advanced significantly with Optimus, its humanoid robot.
I came across two really fascinating movies (here and here) from a robotics specialist that emphasize how crucial a humanoid robot’s speed is in a manufacturing environment.
Although Optimus can reach about 2 mph at the moment, the expert recommends 3 mph as the lowest feasible speed for factory use. Optimus’s main obstacle is still fine-tuning the actuators that drive joint movement,
but if this obstacle is cleared, production should pick up speed and significant advancements should be made this year.
Other forecasts include the possibility that Tesla would create several dozen Optimus robots by 2024. Although Elon Musk and Tesla have set some very lofty goals,
it’s important to highlight Tesla’s development into a business that offers more than simply electric vehicles. Although we already knew that Tesla was a manufacturer and energy storage firm,
this year will see them swiftly establish themselves as a robotics company as well. In 2024, other robotics businesses may release products, but Tesla’s Optimus is expected to evolve at the fastest rate.
4-Acceleration using Open Source
My next prediction: open-source large language models will accelerate in performance. This is the thing I’m most excited about this year. In the video below,
I showcase a trend curve highlighting the closing gap between open and closed-source models, predicting that the release of LLaMA 3 will further narrow the difference.
Hugging Face currently includes over 325,000 open-source AI models, and I see this site growing even more in the future with a focus on how crucial it is to advance quantization methods.
Another reason I think the Mixture-of-Experts will become the de facto norm for open-source models is that it makes large models perform very well because inference doesn’t need using the full model.
Thus, even though it’s a big model, consumer hardware can handle it. The interesting part comes when quantization techniques are introduced.
Meta is still leading the charge when it comes to open-source AI, with Apple also getting into the game, not to mention NASA and IBM with geospatial data.
We should also expect to see a lot of industry-specific models this year, such as Finn GPT and Finance – some that might actually be completely new models themselves.
Apple’s release of their multimodal model ML Faret last year, which mostly flew under the radar, is completely open-source. This – along with the
amount of software that’s come out recently that’s able to leverage the power of Apple Silicon – leads me to believe that Apple wants AI and LLMs to live on device.
In Jaron Lanier’s counterargument below, John warns against the fallacy that open-source software causes decentralization because of network effects.
According to Lanier, the sharing of free resources could lead to centralization akin to a monopoly, and businesses might safeguard their data, making it more difficult for open-source models to obtain datasets.
One way to get around this problem would be to use synthetic data, which I discuss in number 7 below.
5 – AI Agents
Next up, I think AI will usher in the era of AI agents. This year, I saw a lot of advances brought forth by sophisticated models and improved teamwork.
AI agents will find practical uses as they create templates for coding, research, and dataset production use cases. Emerging AI agent behavior,
shown in relationships and habits, will gain prominence and prompt inquiries about the nature of consciousness and the human condition.
AI bots should help with outcome prediction in a variety of circumstances, including testing game theory scenarios such as the prisoners’ dilemma,
in my opinion. These agents have the potential to significantly advance social science, provide an accurate representation of human behavior, and forecast future community behaviors.
Artificial intelligence (AI) agents may find use in anything from political polls and advertising to psychology, psychiatry, and dating. They might act as predictive instruments,
enabling scalable experimentation and doing away with the necessity for human trials. In 2024, I also see more tools to help in managing AI teams.
The most challenging aspect of managing AI agent teams at the moment is not so much putting them into practice as it is figuring out how to define responsibilities, prompts, system messages,
and other elements so that the teams work well together and produce the desired results. I’ll be reviewing a new AI agent project soon, so stay tuned.
6 – No AGI
Next, I believe that in 2024 artificial general intelligence (AGI) will not exist. In the video below, Mark Zuckerberg and Lex Friedman talk about the idea of superintelligence.
They both recognize that there has been progress in the last year, but they are unsure about the date on which a single AI system with general intelligence would be developed.
Zuckerberg highlighted the existence of institutions and organizations that already demonstrate intelligence higher than that of a person, such as the stock market and businesses with a single brand that function as a unit.
Although AGI is clearly desired by the public, as demonstrated by the overwhelming response to Sam Altman’s recent X poll, he anticipated that it would not be feasible by this year.
It will be interesting to watch if Ray Kurle and Elon Musk’s forecasts come true, as they appear to believe that artificial intelligence will arrive in 2029.
7 – Synthetic Data
Next, given the increasing value and difficulty of obtaining data, I believe that 2024 will be the year of synthetic data. Synthetic data is a solution
that involves huge language models in creating fresh data for training future models, as companies protect their precious datasets. I see a
growing trend in the use of synthetic data fabrication for model training, even though its efficacy is still up for debate.
Tesla’s edge in fully autonomous technology is a prime illustration of this, as it has accumulated a significant amount of real-world data from its vehicles’ cameras over the years.
On the other hand, obtaining said data presents difficulties for other automakers. They can install cameras and gather data from the real environment, buy third-party datasets (which are hard to come by), or create artificial data.
In sectors like healthcare and finance where privacy is a major concern, synthetic data is anticipated to be vital. In many industries, employing
user data is risky from a legal standpoint, hence synthetic data is a desirable substitute. In order to build Artificial General Intelligence (AGI),
it is imperative to solve the synthetic data dilemma because firms are unwilling to share their valuable datasets and humans may not generate enough data manually.
8 – Multi-Modal
After that, I think we’ll stop using text-only models and start using multi-modal models by default in 2024. This trend is demonstrated by models such as Gemini,
upcoming iterations of GPT-4, and Apple’s ML Faret. These models are going to be trained with a wide range of inputs, such as text, audio, video, and photos.
Multimodal will not be without difficulties, though. The Standford lecture video below addresses a number of topics,
including the dominance of one modality over others, noise introduced by other modalities, and partial coverage of particular modalities in particular datasets.
Still, I am bullish about the potential of multi-modal capabilities and expect big things in this space by 2024.
9 – Evil Bots
Detecting bots will become nearly impossible in 2024, and synthetic data could be a key contributor to this challenge. Data proliferation
improves models and malicious bots, particularly those designed for scams and spam. Bots are already proficient at solving CAPTCHAs and other
“prove you’re human” tests, making distinguishing between AI and human users difficult.
Elon Musk suggests charging a nominal monthly fee to use platforms like X in order to fight bots. The idea behind this strategy is to make it
economically impossible to maintain a bot network because it becomes very expensive to run multiple bots at once.
The speaker contends that by erecting a financial barrier, it upsets the bot networks’ incentive structure and reduces their profitability for bad actors.
Given the prevalence of deepfakes, the difficulty in identifying bots, and other deceptive tactics in the upcoming 2024 election cycle,
we will need to place a strong emphasis on education to enable people to access online information and assess its reliability.
To enable people to decide for themselves whether or not to trust what they read, this education is an essential step.
10 – GPT 4.5
Last but not least, I anticipate that GPT 4.5 will be released in Q1 or Q2 of 2024 and that it will represent a significant evolution in a number of areas.
But since Sam Altman has apparently said that they are not working on GPT 5, it won’t be a leap to GPT 5. Though GPT 4.5 will still be built on the GPT 4 model and architecture,
I believe it will provide improvements in performance, speed, and cost-effectiveness.
Many reports surfaced toward the end of 2023 suggesting that GPT 4.5 was already in use, but it turned out that these were the product of AI conjecture and delusion.
It appears that some people used artificial intelligence (AI) to fabricate remarks that claimed GPT 4.5 was used, even though this was untrue.
2024 is going to be an interesting year, and I can’t wait to share with you a number of tutorials, reviews, and conversations on more general AI subjects. Kindly share your opinions with me in the comments area.
