The most important AI trends in 2024
ฝัง
- เผยแพร่เมื่อ 8 พ.ค. 2024
- Explore IBM watsonx → ibm.biz/get_started-watsonx
AI is growing at a frenetic pace - just look at how things changed over the course of 2023!
So, what AI developments should we expect in 2024? In this video, IBM Master Inventor Martin Keen discusses this very topic and offers nine trends to watch out for in 2024 ... and possibly a 10th?
00:24 - Reality Check
01:22 - Multimodal AI
01:29 - Smaller Models
04:10 - GPU + Cloud Costs
04:56 - Model Optimization
06:03 - Custom Local Models
06:47 - Virtual Agents
07:14 - Regulation
07:39 - Shadow AI
Get started for free on IBM Cloud → ibm.biz/sign-up-now
Subscribe to see more videos like this in the future → ibm.biz/subscribe-now
#ai #trends #llm #ml
1.Reality check
2.Multimodal AI
3.Smaller model
4.GPU + cloud costs
5.Model optimisation
6.Custom local models
7.Virtual agents
8.Regulation
9.Shadow AI
10.!? guess
Thank you, his writing was drunk chicken scratch...
if one is able to look into building smaller and much more customised local models; that could be a really really cool business.
AI in the future, won't live on clouds. they will live on local servers.
Airbnb for gpus
I like listening to this guy. Soothing, and subtly humorous
My view too !
He's my favorite presenter.
Trend #10 - AI Red Team + Cybersecurity AI.
Totally agree with you with one condition which is Cybersecurity AI is all time trend, not only for 2024
Generative cybersecurity?
@@virtualinfluencer I would rather think of security measures to keep up with prompt hacking techniques. The more you use LLMs in comercial services the more important it will be to prevent abuse and data breaches. Also with increasing capabilities of autonomous agents you probably have to improve or rethink measures suc as captchas.
Exactly, and this should actually be on top of the list. Weird that sth as vague as "shadow ai" was put on the list but ai cybersec wasn't.
I have been working on lots of the items you listed for a awhile. From my perspective, your video was spot on. Love your videos!
Great video. Trends are spot on. I would add "Better Memory Algorithms" as the 10th. Current algorithms have serious limitations as we journey towards AGI.
I love your entire series on these topics. So informative, nicely presented.
Local add-in cards, PCI LPU's or the like (think local Groq)... and/or GPU VRAM will trend upwards in the short term or specialty graphics cards meant for handling small to mid sized LLM's will be released . I think the value of having local models being more accessible with some hardware development is inevitable. I know the cloud argument; but having the speed, affordability and security of local models is still a compelling consideration. Another great video Martin!
I always find unique information with crisp analysis here. The concluding part of the video is very interesting. 👏
This man is the best, I mean all of them on IBM are good but he is like that teacher we loved to go to the class of
TREND #10 - A renewed focus on data quality and Integrity with controls in place to manage. So I am a data guy so I would say this. For me trend 10 is the realisation to business that in order to create and utilise AI as required, their data quality needs to be MAX level. IT world seems to have moved away from ISO etc in recent years (namely due to IaaS, PaaS, Serverless etc.) . Frankley many org's data quality is poor. Good AI requires high quality 'meaningful information'. If you cannot trust your AI 100% of the time, then it is a guide
The focus on multimodal AI is fascinating and truly a game-changer!
Great presentation. Super clear. The speaker's enthusiasm was infectious!
I love his videos! He is entertaining and the information is to the point. I think an important trend will be in saving electrical energy by a smarter network and consumer management.
Francisco
Here are a couple of Trends#10:
1. Commercial competition on AI alternatives
2. Lies and fake detection, good and bad data discrimination
I'm worry about the second option mostly
10th trends. (Actually, there are several.). 10.1 - Larger context windows. 10.2 - Faster ASIC hardware, esp. on the inference side, like Groq. (See investments in these area.)
Can't wait to see the new Virtual Agents, and further work on Foundation Models
This guy is a real genius! I have rarely met smart people like that, hats off!
Trend #10: training in AI as a popular topic. Everyone will try to adquiere skills in the area.
Good video. One point to add to the smaller model category would be the prospective use of bit-sized LLMs like the bitnet along with MoE. Would be interesting to see how well they can perform compared to MoE and if it's effective, could be a game changer. Earnestly hope work in that field will be fruitful.
Very Nicely done - keep going IBM!
You sort of covered it with small models and optimisation. But that appeared to be more about training time and energy. The specific tenth trend is edge AI. Models running entirely on users own device and not a separate cloud server which returns the results
Trends #10 - #20 - back to the roots. AI is so much more than the generative variant. Models like regression, logreg, svm, random forests/boosted trees, and also time series etc have a proven value proposition, we know how they work and they are easy to understand. In fact the EU AI Act implicitely mandates these simpler models for many applications due to requirements of transparency and the right to a human judgement.
trend 10:
emphasis on security and privacy - businesses will want warranties of data confidentiality, deciding between on-device processing, self-hosting servers, vs. cloud
and on the opposite side, AI companies will have to proceed very cautiously with open-sourcing things, because (intentional) mismatch of models and datasets may result in huge harm, with legal disputes on responsibility
#10 Open Source AI
10th Specialized autonomous agents.
11th desktop models, or even models that run on iPhones.
My suggestion for the tenth AI trend that SHOULD be happening in 2024 is AIs seeking the objective truth. That is, rational conclusions drawn from verified facts and sound logic. Currently AIs are sweeping up human information that is steeped in lies and distortions. They're responding to our queries with the falsehoods - about climate, political power, economic rationality, etc. - that we ourselves have placed there. What do you think, Dr. Keen?
Bias will be inherent in any data. There will never be a pure logical, unbiased data. So, models will always have some bias. Our goal is to reduce bias to an acceptable extent.
Benchmarks -- Alignment
Agents working with agents, all org chart with agents, starting from CEO, i think thats most promising direction to AGI 🚀
10th trend: integration with robotics
This
Definitely
Security alignment - for specific industry?
Another great video, thanks for these insights!
Wow, More intresting things developing around.
standardization of data representation for interoperability, different a.i. check the answers to verify allucinations, ai driven robot to introduce reality/ timeline/tasks/physic word, better multimodal transformers, and not this year but fotonics could be game changer for power consumption.
Thanks for this video
I think trend number 10 is actually a very great extension of trend 7. It's not just about virtual assistants. It is also about efficient integration with OS and applications (including those without API). Something like boosted Robotic Process Automation driven by GenAI without prior coding of the software robot.
❤
👍
10) Models learning to consume APIs and interact with software like us (Performing CRUDs on systems) -> this will also assist in voice based UI which is key for XR(VR/AR) systems success, replacing dependency from keyboards, mouse and touch screens for input.
Models are not very good at GUI, as us, but they don't need GUI as us, to perform all these actions, A CLI base system will do the work as well.
#10 evolving the cycle relationship between humans and machines
What about multi agent frameworks? That stuff is absolutely amazing
The tenth requirement for AI is conflict resolution. One huge human short coming is opinion conflict. We battle over religion, government policy, economic policy and truthfulness of information. All information needs to be judged on the relative harm imposed by the use of information to inform decisions by individuals and organized groups.
Here are a couple more candidates for Trend #10…
1. Vector libraries (like Pinecone) for model memory
2. AI Ops positions at regular companies for integration of practical AI solutions in everyday workflows
Hi there. I'm here from 2025 and wanted to let you know that this was way off. 2024 was a much more impressive year for tech than any of us could have imagined!
We can see when you posted this 😂
Quick Summary 🙂
1. Reality check - on how overall AI is being used for innovation as the hype settles.
2. Multimodal AI - models that can treat diverse tasks - e.g. ask about image and get answers in language.
3. Smaller Models - are starting to perform better on many benchmarks compared to their “big” brother - billions of parameters performing better than trillions of parameters.
4. GPU & Cloud Costs - the less consuming resources and more problem specific reduce costs for customers.
5. Model Optimization - techniques such as quantization helping speed-up training; more should be coming.
6. Custom Local Model - finetuning on an organization’s specific needs using its data.
7. Virtual agents - task automation - more applications to come/to be identified for innovation.
8. Regulations - EU acted AI act in in recent time, more to come.
9. Shadow AI - used by internal resources for informal uses in an organization.
10. **** decide on one using your observation ****
Did you have AI edit the earlier mosst important blockchain trends video using different nouns?
I think you did! Very clever.
Maybe prep one for fusion powerplants, or room temperature superconductors. Just to be ready.
Preparing Data which LLMs can understand. Either for RAG or for fine tuning. Ex: If your product documentation is badly written and unstructured, and have complex charts, tables etc📊, then LLM will give poor results.
#10. Enterprise use
We'll be hearing more and more companies tout their use of AI in a more real sense. And it's like the crypto trend of the last few years. Companies that can prove they are benefiting from AI will immediately see their stock prices appreciate
Multimodal AI seems an interesting dimension. Definitely something as an area of interest for innovative ideas.
My vote for trend number 10: AI source classification : Is an object (tile sound movie) partly or complete AI generated?
Permission to use AI at work, permission to breath, permission to exist, perfect
Candidate for #10: Lawsuits and policy arguments over copyright infringements stemming from training data collection.
Trend 10 --- more and more integration of LLM to motorized Robot like Figure AI 01.
Maybe #10 - embodied AI.
Not on a consumer level but still.
The study by Ernest and Young regarding Shadow AI - is that published yet? I'd like to review it.
01:12 2024: The year of more realistic AI expectations
02:24 Generative AI extends its capabilities in multi-modal AI
03:36 Smaller AI models are gaining traction due to their lower resource consumption.
04:48 Shift towards smaller models for cost-efficiency and local runability.
06:00 Model optimization is a key trend in 2024.
07:12 Model optimization techniques will emerge in 2024
08:24 Regulation and shadow AI will impact AI trends in 2024
09:35 The dangers of generative AI are increasing along with its capabilities.
#10 - AI in schools, 1) I'm thinking it's potentially the most significant way of teaching language skills by having a voice chat partner fluent in all languages. I think that's why we're seeing Rosetta & Babbel being pushed in marketing blitz atm, they know the clock is ticking before they are irrelevant. 2) Coding assistants in schools 3) teaching kids to be early adopters is an inevitability and OpenAI is testing giving students access to 4-turbo for free for this very reason.
Trend #10 me in the 2024 AI "System"
Trend #10 will be an intensified focus on developing the next renewable energy source that will power the jump to quantum computing. Current energy source / power consumption limitations coupled with current computing power limitations combine for a ceiling we seem closer to hitting than is talked about broadly. The next big jump in AI innovation will directly coincide with unlocking the ability to sustain quantum computing solutions the way we sustain current processing speeds present day. That means solving for the power needed to keep next gen systems processing indefinitely in a cost effective and scaleable manner.
Interesting video
Basically, everyone is trying to reduce dependence on NVDA? lol
Well when I was thinking this out 20 years ago the next big trend is light memory --- so that instead of magnetic arrays you have light imprints a bit like CD but on light processors so instead of having two states you have frequency states. memory itself can also self recycle in loopback arrays to make lightspeed processing with little or n heat gain due to generation. It really helps the ai in processing complex memory tasks when memory is condensed algoryhtmically. "Also time based ai" so that data processes and training get marked in time code as well as physical relative spatial connections to enable it to physical rules processing that is adaptive based on functioning correctness and goal obtainment. So information isn't based simply on last information but rather temporal forecasting of states of correctness. This then is also affective on the light arrays for optimized processing states.
The idea of time-based AI makes a lot of sense. It is a little like looking at StackOverflow for programming answers without noticing that some answers are from this year and others from 12 years ago. Time has a degrading impact on the relevance that doesn't seem to be accounted for. It may not matter much in generative images or prose, but it is very important with changing technologies and with real-life political arenas, for example.
10: A reversion to staff hiring after all the AI related layoffs when it is finally realised that AI is not ready for prime time either via unsustainable compute costs, cyber hacks or legal woes.
New services / agencies coming up, for setting up the AI models for businesses
Number 10 - Cybersecurity and Blockchain A.I
Gpt 5 soon 🔥🔥🔥
10 - Personal assistants with vector databases.
Airbnb for GPUs is trend #10. Basically folks will join a decentralized protocol to rent their spare GPU capacity over the network to run low bandwidth training workloads.
Wait wait wait...Homebrewing isn't your full job! Who knew that my love of beer and tech would collide like this!
(Presenter has an excellent channel called 'The Homebrew Challenge')
Adoption and Change Management in Enterprises and Organizations, from upskilling to creating champions
#10 - Proliferation of end to end Apps will start challenging established business software players
10th should be enviromental friendly. How much it can save energy with High computing usage in Deep learning and Generative AI?
What is missing: Automating hallucination detection and correction before output
Amazing video
In my view, efforts to decrease the BIAS in trained models and datasets is one and a crucial trend. if not , the deviation caused by these biased models is going to destroy many benefits imagined. MODELS that can detect biased results are needed.
Emergence of breakthrough layer 2 applications that organisations can adopt to execute end-to-end processes
I think the 10th trend will be the rise of a highly competitive market for AI tools - well-packaged service offerings and alternatives - remember iPhone dominance initially and then the explosive smart phone market? Expect the start of targeted marketing, custom offerings, and aggressive GPT4 alternatives in price and packaging.
I just took an interview with IBM, amazing company, hit me up any buddy like IBM
The 10th trend: the militarization of AI in analyzing battle field data, strategy, and weaponry.
10th increassing impact from good and bad data, which is feeding the diffrent LLM
For trend # 10, Devin says hellooooo
10. Evaluation
This is a good one! Yes, evaluation - self evaluation on data points, but also redefining suggestion of the input (more like a sophisticated "do you mean [...]?")
I agree. I think that evaluation/critique holds the most potential for increasing the quality of output.
10. Hopefully less screen time so we have more physical activity
#10 - Competition between the tech giants to build the smartest LLMs, on a race towards AGI
10. Federated Learning - Training AI models while preserving data privacy. The answer to the EU‘s AI act.
10. Knowledge Graph
Have to say, as an IT person of some 35 years, I am over the hype each time we develop a new tech, and this is no different to AI. The hype generator ultimately is fuelled by people looking to make a buck or two - an example is the TH-camr Wes Rotham (?) ... to many titles driven by the desire to standout without "SHOCKING" with its headline. Its a shame really, as some of the content from Wes is really good - but I can't be bothered to sift through the BS.
This series by big blue, and this speaker, on the other hand are more levelled and trustworthy imo. And the content is equally as good imo. As a result, I definitely recommend, and will continue to recommend to those seeking enlightenment without the hype BS.
Thank-you for the content.
10. What about DeepFake, Deep voice cloning? I was amazed by a couple of new AI companies this year 2024 that do custom voice cloning which can then be used to replicate old songs with my own voice perfectly without using karaoke. Additionally, this also replaces the voice or it can merge voices which will give a different output which I was amazed and I don't think we have this before that two voices can be merged with some amount of percentage of one voice merge with some amount of percentage of other voice which is super cool.
I think on-device learning will gain momentum
I think the biggest trend this year will be more apps designed to make these things useful for regular people, without all these dumb hoops most services make you jump through, or having to know how to use obscure phrasing / "prompt engineering".
10. Models based on better architecture
good video!
10. Increased productivity
AI cybersecurity will be another trend from my observations. The compounding and increased complexity of AI cyber attacks from unsavory individuals, or groups will demand a drastic increase in AI Black Hats and cyber security professionals. This industry will drastically increase due to a vast influx of data from users utilizing AI models of multiple scopes and uses. This will massively increase the need for data storage, computing resources, and therefor a more robust set of regulations, technologies, and institutions to manage the demand.
For the 10th trend I would expect either of these three to play a vital role
- Multi-agent generative systems, training a "workforce of highly specialized AI agents"
- AI for (cyber) security
- RAGs (Retrieval-augmented generation)
10. Custom hardware for acceleration :)
First, how does one do this- writing in front of character/ anchor with no presence of glass/ board seen? His writing too is front facing, meaning he is writing in reverse direction with mirror image of letters! Somebody please explain me. Second, I did not go beyond three-year it seemed too broad and yes somewhat technical.
He practiced writing backwards dilligently.
Is no/low code AI a latest trend for 2024
Embedded and local AI in customer devices
running local model for maximum privacy
#10 AGI
A nice trend would be an AI as a law adviser. 🙈🙉🙊
the 10th may be an AI model with a 'backspace' token which allow it not forced to ’deductive reasoning‘ a mistaken starting word
that said, it's really hard to train AI with 'human talking data' instead of 'human talked data'
My question is; how IBM is helping developing countries to catch up with the technology gap that exists between well-off countries and the poorer ones?.
Democratising LLMs. Make it available on the edge for example