
Just as there are widely understood empirical laws of nature - for example, what goes up must come down, or every action has an equal and opposite reaction - the field of AI was long defined by a single idea: that more compute, more training data and more parameters makes a better AI model.
However, AI has since grown to need three distinct laws that describe how applying compute resources in different ways impacts model performance. Together, these AI scaling laws - pretraining scaling, post-training scaling and test-time scaling, also called long thinking - reflect how the field has evolved with techniques to use additional compute in a wide variety of increasingly complex AI use cases.
The recent rise of test-time scaling - applying more compute at inference time to improve accuracy - has enabled AI reasoning models, a new class of large language models (LLMs) that perform multiple inference passes to work through complex problems, while describing the steps required to solve a task. Test-time scaling requires intensive amounts of computational resources to support AI reasoning, which will drive further demand for accelerated computing.
What Is Pretraining Scaling? Pretraining scaling is the original law of AI development. It demonstrated that by increasing training dataset size, model parameter count and computational resources, developers could expect predictable improvements in model intelligence and accuracy.
Each of these three elements - data, model size, compute - is interrelated. Per the pretraining scaling law, outlined in this research paper, when larger models are fed with more data, the overall performance of the models improves. To make this feasible, developers must scale up their compute - creating the need for powerful accelerated computing resources to run those larger training workloads.
This principle of pretraining scaling led to large models that achieved groundbreaking capabilities. It also spurred major innovations in model architecture, including the rise of billion- and trillion-parameter transformer models, mixture of experts models and new distributed training techniques - all demanding significant compute.
And the relevance of the pretraining scaling law continues - as humans continue to produce growing amounts of multimodal data, this trove of text, images, audio, video and sensor information will be used to train powerful future AI models.
Pretraining scaling is the foundational principle of AI development, linking the size of models, datasets and compute to AI gains. Mixture of experts, depicted above, is a popular model architecture for AI training. What Is Post-Training Scaling? Pretraining a large foundation model isn't for everyone - it takes significant investment, skilled experts and datasets. But once an organization pretrains and releases a model, they lower the barrier to AI adoption by enabling others to use their pretrained model as a foundation to adapt for their own applications.
This post-training process drives additional cumulative demand for accelerated computing across enterprises and the broader developer community. Popular open-source models can have hundreds or thousands of derivative models, trained across numerous domains.
Developing this ecosystem of derivative models for a variety of use cases could take around 30x more compute than pretraining the original foundation model.
Developing this ecosystem of derivative models for a variety of use cases could take around 30x more compute than pretraining the original foundation model.
Post-training techniques can further improve a model's specificity and relevance for an organization's desired use case. While pretraining is like sending an AI model to school to learn foundational skills, post-training enhances the model with skills applicable to its intended job. An LLM, for example, could be post-trained to tackle a task like sentiment analysis or translation - or understand the jargon of a specific domain, like healthcare or law.
The post-training scaling law posits that a pretrained model's performance can further improve - in computational efficiency, accuracy or domain specificity - using techniques including fine-tuning, pruning, quantization, distillation, reinforcement learning and synthetic data augmentation.
Fine-tuning uses additional training data to tailor an AI model for specific domains and applications. This can be done using an organization's internal datasets, or with pairs of sample model input and outputs.
Distillation requires a pair of AI models: a large, complex teacher model and a lightweight student model. In the most common distillation technique, called offline distillation, the student model learns to mimic the outputs of a pretrained teacher model.
Reinforcement learning, or RL, is a machine learning technique that uses a reward model to train an agent to make decisions that align with a specific use case. The agent aims to make decisions that maximize cumulative rewards over time as it interacts with an environment - for example, a chatbot LLM that is positively reinforced by thumbs up reactions from users. This technique is known as reinforcement learning from human feedback (RLHF). Another, newer technique, reinforcement learning from AI feedback (RLAIF), instead uses feedback from AI models to guide the learning process, streamlining post-training efforts.
Best-of-n sampling generates multiple outputs from a language model and selects the one with the highest reward score based on a reward model. It's often used to improve an AI's outputs without modifying model parameters, offering an alternative to fine-tuning with reinforcement learning.
Search methods explore a range of potential decision paths before selecting a final output. This post-training technique can iteratively improve the model's responses
Most recent headlines
05/01/2027
Worlds first 802.15.4ab-UWB chip verified by Calterah and Rohde & Schwarz to be ...
01/06/2026
January 6 2026, 05:30 (PST) Dolby Sets the New Standard for Premium Entertainment at CES 2026
Throughout the week, Dolby brings to life the latest innovatio...
02/05/2026
Dalet, a leading technology and service provider for media-rich organizations, t...
01/05/2026
January 5 2026, 18:30 (PST) NBCUniversal's Peacock to Be First Streamer to ...
01/04/2026
January 4 2026, 18:00 (PST) DOLBY AND DOUYIN EMPOWER THE NEXT GENERATON OF CREATORS WITH DOLBY VISION
Douyin Users Can Now Create And Share Videos With Stun...
13/03/2026
Share
Copy link
Facebook
X
Linkedin
Bluesky
Email...
13/03/2026
Harmonic (NASDAQ: HLIT) today announced Spectrum X Plus, the newest generation of its Spectrum X media server, offering double the channel density of previous ...
13/03/2026
When Ewing Covenant Church made the decision to return to its original, historic building, affectionately called 1867 Sanctuary for weekly worship, the congre...
13/03/2026
Marshall Electronics introduces its first all-IP 4K POV camera, the CV574-WP, at NAB 2026 (Booth C8339). The CV574-WP supports NDI |HX, providing ultra-efficien...
13/03/2026
At NAB Show 2026, Net Insight introduces Nimbra 520, a high-density media processing node designed to simplify live contribution and distribution across both ma...
13/03/2026
Abandon Editorial Signs With Micha l Dimitri for West Coast Representation
Brie Clayton March 12, 2026
0 Comments
Abandon Editorial is excited to part...
13/03/2026
Documentary The Bulldogs Shot and Edited with Blackmagic Design
Brie Clayton March 12, 2026
0 Comments
Editorial tools helped shape film in real time,...
13/03/2026
AE Captions as Fast as CapCut - No Plugins
Graham Quince March 12, 2026
0 Comments
Stop wasting hours clicking through nested compositions and manuall...
13/03/2026
New Music USA and Berklee Institute of Jazz and Gender Justice Announce 2026 Nex...
13/03/2026
RT has announced today that Rick O'Shea is the new presenter of Arena RT Radio 1's flagship weeknight arts and culture programme. Rick has been pres...
13/03/2026
Lights! Camera! Action! The 98th Oscars set to air live as RT backs the Irish n...
12/03/2026
Staines-upon-Thames, UK, 11th March, 2026 - Yospace, the trusted leader in Dyna...
12/03/2026
Utah Scientific Expands Technology Partner Program With Integrations From Audina...
12/03/2026
Techex, a global expert in live video solutions over IP and cloud, announces the appointment of Matt McKee as Senior Director, Sales, Americas, further strength...
12/03/2026
KOKUSAI DENKI Electric America has appointed Mondae Hott as Regional Sales Manag...
12/03/2026
At the 2026 NAB Show, Interra Systems will showcase its latest advancements in a...
12/03/2026
The 15th National Games of China concluded after a two-week celebration of athletic excellence and regional collaboration. Held from Nov. 9-21 across Guangdong,...
12/03/2026
Live-production academic program Butler Sports Live produced a total of 40 fall-...
12/03/2026
The University of Nebraska's HuskerVision has completed the second phase of ...
12/03/2026
Grass Valley and integration partner Tab M Solutions have completed Phase 1 of a...
12/03/2026
The broadcaster expands its campus-production model as the university handles tw...
12/03/2026
Disney has announced the addition of March Madness - the NCAA Division I Men...
12/03/2026
Apple TV's Friday Night Baseball MLB doubleheader series returns for its f...
12/03/2026
The senior from New Jersey is making his mark in South Bend, both on the mic and behind it...
12/03/2026
After a relatively quiet January, the month of February was jammed packed with l...
12/03/2026
Long-time production partner Echo Entertainment is producing the broadcast, while Cosm played a vital role in the collaboration...
12/03/2026
By Jessica Herndon
We love kicking off each year by introducing the world to po...
12/03/2026
Samrat Chakrabarti, George Basil, Kiran Deol, Katie McCuen and Vishal Vijayakumar attend the 2025 Sundance Film Festival premiere of Didn't Die at the Lib...
12/03/2026
In Latin America, women are shaping music and defining its future. To kick off t...
12/03/2026
En Am rica Latina, las mujeres est n moldeando la m sica y definiendo su futuro....
12/03/2026
Let's turn back the clock 20 years: The music landscape was a world away fro...
12/03/2026
Bad Bunny is no stranger to Spotify's Billions Club. In fact, he has a whopp...
12/03/2026
Spotify was at the London Book Fair this week, joining conversations across the publishing industry about how people can make reading part of their daily lives....
12/03/2026
Mastering tool improves mono compatibility
Tokyo Dawn Labs' Ohlhorst Digital range is a series of mastering-focused plug-ins developed by Jan Ohlhorst, ...
12/03/2026
Wave FX processor integrated into four products
Lewitt have teamed up with Elgato to create a new processor for the company's Wave Next product range, i...
12/03/2026
Free tool for annotating audio files
Mix Notes is a new, free iOS App that provides users with a simple way to annotate their audio files. It's been cre...
12/03/2026
Side-chain ducking tool gets an upgrade
Devious Machines' popular side-chaining and envelope-shaping tool has just been kitted out with an improved enve...
12/03/2026
Ceremony to take place on 16 April 2026
The MPG (Music Producers Guild) have revealed the full shortlist for this year's MPG Awards, which will be takin...
12/03/2026
Emulates three classic dbx 160 variants
The latest arrival to Overloud's Gem Series plug-in range faithfully recreates not one, but three versions of th...
12/03/2026
New granular soft synth announced
Said to be their most advanced software synthesizer to date, Baby Audio's latest release has been built on a new granu...
12/03/2026
Latest version now live!
Edit 11 March 2026 - Bitwig Studio 6 is now live, and available for all to download!
The latest version of Bitwig's DAW softwa...
12/03/2026
Latest free eBook now available!
Designed for recording engineers, audio-technology students and technically minded musicians, our latest free eBook deliver...
12/03/2026
AFL and NITV partner to launch new First Nations led program Inside the Huddle&...
12/03/2026
Rohde & Schwarz Cybersecurity expands SITLine network encryptor portfolio - more...
12/03/2026
Rohde & Schwarz to showcase future-proof EMC testing solutions at EMV 2026 Rohde & Schwarz will participate in EMV 2026, Europe's premier trade fair and c...