- Everyday AI
- Posts
- Will OpenAI run away in the LLM race in 2025?
Will OpenAI run away in the LLM race in 2025?
LLM race in 2025, NVIDIA unveils AI supercomputer, Google working on āworld models,' Samsungās Ballie robot to launch this year and more!
Everyday AI
January 07, 2025
š Subscribe Here | š£ Hire Us To Speak | š¤ Partner with Us | š¤ Grow with GenAI
Sponsored by
Looking to transform your org with AI? Simply giving employees access wonāt cut it.
āThatās like thinking that if we put a treadmill in every home, weāre going to cure heart disease,ā says Conor Grennan, Chief AI Architect at the NYU Stern School of Business.
On the latest episode of the WorkLab podcast from Microsoft, he explains the deeper mindset shift AI requires.
It's available wherever you get your podcasts.
Outsmart The Future
Today in Everyday AI
7 minute read
š Daily Podcast Episode: Will OpenAI run away with the LLM race this year or should they be worried about the competition? We take a look at the LLM landscape in 2025. Give it a listen.
šµļøāāļø Fresh Finds: NVIDIA announces new GPUs, Apple speaks on AI mishap and Razerās plans for an AI gaming copilot. Read on for Fresh Finds.
š Byte Sized Daily AI News: NVIDIA unveils AI supercomputer, Google working on āworld modelsā and Samsungās Ballie robot to launch this year. For that and more, read on for Byte Sized News.
š AI In 5: Weāre revealing 3 secret ChatGPT features only available on the desktop app. See it here
š§ Learn & Leveraging AI: Wondering who will lead the LLM race in 2025? We take a look at OpenAIās recent moves and compare it to the other major players. Keep reading for that!
ā©ļø Donāt miss out: Did you miss our last newsletter? We talked about AGI in 2025, OpenAI losing money on ChatGPT Pro plan, Intelās new AI chip lineup and Microsoft Copilot added to LG and Samsung TVs. Check it here!
Will OpenAI run away in the LLM race in 2025? š
Did Google dethrone OpenAI in December?
Could Claude finally catch up?
Or will ChatGPT reign supreme in the LLM race of 2025?
We break down the LLM landscape in 2025.
Join the conversation and ask Jordan questions on AI here.
Also on the pod today:
ā¢ LLMs and Internet Connectivity š
ā¢ LLM Predictions and Outcomes š®
ā¢ Popularity of Different AI Systems š¤
Itāll be worth your 52 minutes:
Listen on our site:
Click to listen
Subscribe and listen on your favorite podcast platform
Listen on:
Upcoming Everyday AI Livestreams
Wednesday, January 7th at 7:30 am CST ā¬ļø
Hereās our favorite AI finds from across the web:
New AI Tool Spotlight ā Wegic is your AI website team, Jellypod creates customizable AI podcasts and X-Design transforms your product visuals.
Apple ā Apple says that itāll clarify AI summaries after a recent BBC headline was botched.
NVIDIA ā At CES 2025, NVIDIA announced its highly anticipated RTX 50-series GPUs, led by the RTX 5090 priced at $1,999.
NVIDIA also revealed that itās helping humanoid robots learn through the Apple Vision Pro.
AI Chips ā AMD has unveiled new AI chips for laptops and desktops.
AI in Media ā Gaming hardware company Razer has announced an AI gaming copilot.
AI Video - Inworld AI, Streamlabs, and NVIDIA are collaborating on a new AI assistant for streamers that offers real-time technical support and serves as a cohost.
Future of Work ā Klarnaās CEO says that AI is developing so quickly that itāll soon be able to do his job.
AI Models - NVIDIA has partnered with Black Forest Labs to create FLUX FP4.
In collaboration with @nvidia we developed FLUX FP4 for lightning-fast inference on the newly announced GeForce RTX 50 Series GPUs. Read more in our announcement blogpost:
@NVIDIA_AI_PC@NVIDIAGeForce
ā Black Forest Labs (@bfl_ml)
4:43 AM ā¢ Jan 7, 2025
1. Google DeepMind Sets Sights on World Models š
Google DeepMind is assembling a new team to develop cutting-edge āworld modelsā aimed at simulating physical environments, led by Tim Brooks, a former OpenAI executive. This initiative comes as the race for artificial general intelligence (AGI) heats up, with DeepMind emphasizing the need for scaling pretraining on video and multimodal data to achieve their ambitious goals.
Brooks has already announced open positions for research engineers and scientists to tackle challenges in training at scale and integrating these models with existing AI systems.
2. NVIDIA Unveils Personal AI Supercomputer Project Digits š„ļø
At CES, NVIDIA announced the upcoming launch of Project Digits, a compact personal AI supercomputer set to hit the market in May. Powered by the innovative GB10 Grace Blackwell Superchip, this desktop marvel can handle AI models with up to 200 billion parameters and starts at $3,000, making advanced AI development more accessible than ever.
With 128GB of unified memory and the ability to link two systems for even greater processing power, this tool is poised to empower data scientists and AI researchers alike. According to NVIDIA CEO Jensen Huang, this leap forward places a powerful AI supercomputer right on the desks of creators.
3. NVIDIA Unveils Cosmos for Physical AI Revolution š¦¾
At CES, NVIDIA revealed Cosmos, a groundbreaking platform designed to accelerate the development of autonomous vehicles and robots through advanced generative models and video processing tools. With an open model license, developers can harness the power of Cosmos to create photorealistic synthetic data, reducing reliance on expensive real-world data capture.
Noteworthy adopters include industry leaders like Uber and XPENG, signaling a significant shift towards democratizing physical AI.
4. Samsungās Ballie Rolls Towards Consumers in 2025 š¤
Samsung's quirky rolling robot, Ballie, is set to hit the market in the first half of 2025, continuing its journey from concept to consumer. After a five-year wait and a redesign aimed at practicality, Ballie showcased its smart home control abilities and interactive features at CES, sparking both curiosity and skepticism among attendees.
While its portable projector capabilities and visual AI impressed during demos, questions linger about its durability and real-world performance. As anticipation builds, the price remains a mystery.
5. Microsoft Bets Big on AI in India š®š³
Microsoft has announced a $3 billion investment aimed at expanding its artificial intelligence and cloud services, along with a commitment to train 10 million people in AI skills. CEO Satya Nadella highlighted the potential of AI in India, noting the country's rapid technological adoption and its position as a key market for U.S. tech giants.
With plans for a fourth data center and partnerships to promote entrepreneurship, this initiative not only boosts India's AI ecosystem but also signals a growing opportunity for individuals and businesses to leverage cutting-edge technologies.
6. Anthropic Eyes $2 Billion Funding Boost š°
Anthropic is reportedly in advanced talks to secure $2 billion in funding, potentially catapulting its valuation to $60 billion, as per the Wall Street Journal. This injection of capital, led by Lightspeed Venture Partners, would solidify Anthropicās position as the fifth-most valuable startup in the U.S., following giants like SpaceX and OpenAI.
With heavy financial backing from Amazon, which has invested a total of $8 billion since 2023, and additional support from Google, the tech landscape is increasingly focused on AI innovations.
3 Secrets ChatGPT Desktop Features Revealed!
Did you know there are ChatGPT capabilities exclusive to the app that you CANāT get on the web version?
Weāre showing you 3 secret features on the ChatGPT desktop app that are available to all users.
Check out today's AI in 5.
š¦¾How You Can Leverage:
OpenAI's reportedly burning $5 billion in cash, yet fully chasing agents, AGI and ASI.
Google's Gemini just had its best month ever after sleeping at the wheel for a few years.
And Claude? Maybe it could rebound in 2025 if it updates its models or at least gives you enough messaging to actually use it.
For two years, OpenAI ran this LLM race kinda solo.
Now Google's finally laced up its shoes and making the Generative AI race more interesting.
Why should you care?
Even if youāre not using LLMs on the front-end, their development literally touches every aspect of your life.
The majority of enterprise software, new devices we use and webpages we browse are being puppeteered by LLMs.
Will OpenAI win?
Today, we gave you 3 reasons why they might not and 3 reasons why they might.
Hereās the gist, shorties.
Why OpenAI Might Take an L š¬
1. The Benchmark Burnout
Remember obsessing over MMLU scores?
That's so 2024.
OpenAI's quietly backing away from the numbers game faster than crypto bros ditched their NFTs. They know something we don't: business value just hits different than test scores.
And with a focus on agents, AGI and ASI, the benchmark jostle might not be the name of the LLM game in 2025.
What it means:
Stop. Making. Decisions. Based. On. Benchmarks. Full stop.
Take your three most expensive business processes and test each model specifically on those. Run a two-week bake-off between GPT-4, Gemini, and Claude. Track completion time, accuracy, and actual costs.
Then make your choice.
2. Search: The Feature That Ate ChatGPT
Their new ChatGPT Search feature?
Ask it a follow-up question and watch it malfunction like it's in a time loop.
Internet connectivity is imperative for any LLM, and we think OpenAI took a massive step backward with their ChatGPT Search functionality.
What it means:
Split your workflow starting tomorrow. Use ChatGPT for your core analysis and writing and heavy LLMing.
Try routing all your research and real-time data needs through other tools, like Googleās Deep Research.
3. The $5B Reality Check
While OpenAI chases the AGI and ASI dragons, their core products could collect dust.
That $200/month Teams plan? Sam Altman literally tweeted they're losing money on it. Not exactly confidence-inspiring.
And recent reports show that OpenAI is losing as much as $5 billion a year. Yikes.
What it means: You gotta follow the money.
Weāll see if OpenAI focuses a bit more on revenue in 2025, or if theyāll just continue to fundraise and let that whole āprofitabilityā thing sort itself out.
Why OpenAI Might Still Run This Town šŖ
1. The Half-Billion User Death Star
The numbers don't lie. ChatGPT has WAY more users than everyone else COMBINED.
Every prompt makes them stronger. Every conversation feeds the beast. Google and Claude are playing catch-up in the AI chatbot game OpenAI defined.
To make matters even worse for their competitors?
Both Apple Intelligence AND Microsoftās Copilot are leveraging OpenAIās models, putting the GPT tech in front of hundreds of millions of users.
More users. More data. More separation.
What it means:
Users give LLMs data. Data makes LLMs better.
The more users, the better the models.
Thatās why OpenAI could still win the 2025 race, even with Google on its heels.
2. The Small Model Surprise
The recent report from Microsoft on model sizes is telling.
Take a look at model size and performance on OpenAIās own models, and how much their models have improved on a per-parameter basis:
GPT-4 = 1.7 trillion parameters, 86.4 MMLU
GPT-4o = 200 billion parameters, 88.7 MMLU
GPT-4o mini = 8 billion parameters, 82 MMLU
Now, letās compare that GPT-4o Mini to other smaller models with known parameter sizes and MMLU scores:
GPT-4o mini = 8 billion parameters, 82 MMLU
Llama 3.2 11B = 11 billion parameters, 73 MMLU
Microsoft Phi-3 7B = 7 billion parameters, 65 MMLU
What it means:
OpenAI's mini models are eating everyone's lunch, on a power-per-parameter basis.
Their 8B parameter GPT-4o mini model is dunking on competitors 3x its size.
Edge AI isn't just coming - OpenAI's already got their tiny titans ready to deploy once the hardware catches up.
(Which is definitely happening this week at CES).
But no oneās talking about it.
We see a future where users are using dozens/hundreds/thousands of small language models at once, that are fine-tuned for SUPER niche verticals.
OpenAI is sneakily already crushing it here. Just no oneās been paying attention.
3. The Features-First Flex
While Google's busy hiding most of their best features in its developer-friendly AI Studio, OpenAI built an empire regular humans can actually use.
They've got reasoning models, GPTs, Canvas, and code generation and a handful of other features and tools all playing nice together.
The interface? Chef's kiss.
What it means:
Claude is sleeping when it comes to model updates and needed-features, though we love some Artifacts.
Gemini FINALLY improved their front-end interface, after probably losing billions/trillions in market cap by putting their worst foot forward to the AI business world.
But OpenAI?
Theyāve been the leader in balancing model power with accessible and useful features.
The verdict?
Will OpenAI win the LLM race in 2025?
Although it wonāt be as close as the last few years, we feel pretty solid in predicting they will.
ā
Numbers to watch
20%
A study by UC Berkeley and Harvard Business School found that high-performing entrepreneurs benefited by over 20% from AI advice.
Now This ā¦
Let us know your thoughts!
Vote to see live results
Do you attend our livestreams?Every weekday, we bring you fresh AI insights, exclusive interviews, and breaking news with our Everyday AI livestream. |
Reply