
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
06/10/2025
France T l visions, France's leading broadcaster, has received the 2025 EBU ...
04/09/2025
Monumental Sports & Entertainment (MSE), in collaboration with Dalet, has been a...
15/06/2025
July 2025 in Dublin, Berlin, Amsterdam & London
Photo: Thea Martre
Music Production for Women (MPW) have announced that they will be running a series of fo...
15/06/2025
Composer/producer launches free virtual instruments
Sulcata Sound is the latest venture of Jason Graves, a two-time British Academy Award-winnning composer,...
14/06/2025
NEW YORK Pluto TV and the All Womens Sports Network have launched a free ad-supported streaming TV (FAST) AWSN channel in the U.S., Canada, the U.K. and the Nor...
14/06/2025
NEW YORK and CINCINNATI E.W. Scripps has announced a new, multiyear agreement with the WNBA that will continue Ions regular-season coverage of the league on Fri...
14/06/2025
WASHINGTON The National Association of Broadcasters highlighted the hidden importance of spectrum in the production of major sporting events and described wha...
14/06/2025
WASHINGTON Sunsetting ATSC 1.0, expanding business opportunities for NextGen Broadcast and increasing international adoption of the ATSC 3.0 standard were top o...
14/06/2025
SAN FRANCISCO Samba TV and Acxiom have announced that they will dramatically expand their longstanding relationship....
14/06/2025
July 2025 in Dublin, Berlin, Amsterdam & London
Photo: Thea Martre
Music Production for Women (MPW) have announced that they will be running a series of fo...
14/06/2025
San Francisco State University's School of Cinema Uses Blackmagic Design
Brie Clayton June 13, 2025
0 Comments
More than 40 Blackmagic Design came...
14/06/2025
Boris FX Mocha Pro Adds New AI Tools To Tackle VFX Tasks Fast
Jessie Electa Petrov June 13, 2025
0 Comments
The 2025.5 release helps artists work more...
14/06/2025
AJA Debuts DRM2-Plus Mini-Converter Frame at InfoComm 2025
Brie Clayton June 13, 2025
0 Comments
Next-gen frame addresses diverse rackmount needs wit...
13/06/2025
(L-R) Lindsay Utz, Michelle Walshe, and The Right Honourable Dame Jacinda Ardern attend the 2025 Sundance Film Festival premiere of Prime Minister at Eccles T...
13/06/2025
Photo credit: Atsushi Nishijima
If you're a true lover of rom-coms, chances...
13/06/2025
Pure Drama and Fierce Rivalries set to dominate the world's most iconic spor...
13/06/2025
Johannesburg, 12 June 2025 - The National Film and Video Foundation (NFVF), an a...
13/06/2025
ABILENE. Texas A severe storm knocked down the tower and severely damaged the news studio and main facility of Sinclair-owned KTXS here on Sunday, June 8....
13/06/2025
Berklee's Music Business/Management Department Recognized by the Music Biz A...
13/06/2025
WASHINGTON The ATSC, the Broadcast Standards Association, honored veteran technologist Aldo Cugnini and Clarence Hau, Senior Vice President of Standards, Policy...
13/06/2025
(Editor's note: The 2025 UFL Championship Game between the D.C. Defenders and Michigan Panthers kicks off Saturday, June 14, at 8 p.m. Eastern. The game wil...
13/06/2025
New iPad/iPhone synth App announced
Following on from last year's release of Gradient Synth - which reached #6 on the App Store's Paid Music charts ...
13/06/2025
LONDON Warner Bros. Discovery has announced that HBO Max will launch direct-to-consumer in multiple new countries this July as the streamer becomes available in...
13/06/2025
AI voice transcription and captioning platform Verbit has added a new feature to its Captivate ASR solution the ability to identify specific features in automat...
13/06/2025
WASHINGTON Federal Communications Commission member Anna Gomez has wrapped up two weeks in California visiting broadcasters, television studio executives, enter...
13/06/2025
WASHINGTON The U.S. House of Representatives voted mostly along party lines to approve a rescission package that would cancel $9.4 billion in previously approve...
13/06/2025
At InfoComm 2025, AJA Video Systems announced DRM2-Plus, an intuitive, high-capacity 3RU frame that can neatly house up to 24 AJA Mini-Converters. Tailored to s...
13/06/2025
Cinema advertising leader to leverage AOS and suite of AI-enabled solutions to optimize forecasting, yield management, and streamlined ad sales and operations a...
13/06/2025
Manfrotto has launched the ONE Hybrid Tripod, a new support system designed specifically for professional content creators working with mirrorless cameras acros...
13/06/2025
Leading video software provider, Synamedia, today announced that its Media Edge Gateway (MEG), an ATSC 3.0 software-based IRD, now supports Device Security requ...
13/06/2025
LiveU, the global leader in live IP-video contribution, production and distribution solutions, is deepening its commitment to the German-speaking market with th...
13/06/2025
Chaos, the leader in architectural visualisation software, today announces Chaos Corona 13, giving archviz designers new ways to add eye-catching style and flai...
13/06/2025
PALI's Nena Music Video Shot with Blackmagic Design
Brie Clayton June 12, 2025
0 Comments
Blackmagic Cinema Camera 6K and DaVinci Resolve Studio b...
13/06/2025
OddBeast Powers Up iRobot's Newest Roombas with Suite of CGI Launch Assets
Brie Clayton June 12, 2025
0 Comments
The motion design and production ...
13/06/2025
On Chick Coreas Birthday, a Newly Uncovered Archival Release The Visitors, composed by Corea and performed by vibraphonist Gary Burton and pianist Kirill Gers...
13/06/2025
In fulfilment of a recommendation by the Government's Expert Advisory Commit...
13/06/2025
SVG Sit-Down: Backblaze's Gleb Budman Talks Products, Partnerships, and the ...
13/06/2025
SVG Sit-Down: DAZN's Walker Jacobs Calls Streaming the FIFA Club World Cup ...
13/06/2025
New Sponsor Spotlight: Vecima Networks' Paul Strickland on How Improving QoE...
13/06/2025
Pitch Perspective: Where's Next for Specialty Cameras in Soccer? Leaders from Sky Austria and ACS discuss the possibilities of camera placement pitchside B...
13/06/2025
Premiership Rugby Final 2025: Vintage clash between Bath and Leicester gets full...
13/06/2025
Premiership Rugby Final 2025: TNT Sports gears up for Bath vs Leicester battle w...
13/06/2025
NCAA Men's College World Series: ESPN Adds Two-Point SupraCam, Invests in Ne...
13/06/2025
New FSWX signal and spectrum analyzer with novel architecture overcomes limits o...
13/06/2025
Apple today announced the addition of iPad to Self Service Repair, providing iPad owners with access to repair manuals, genuine Apple parts, Apple Diagnostics t...
13/06/2025
CUPERTINO, CALIFORNIA Apple today previewed iOS 26, a major update that brings a beautiful new design, intelligent experiences, and improvements to the apps use...
13/06/2025
At Apple's Worldwide Developers Conference (WWDC), Apple unveiled Apple Games, an all-new destination designed to help players jump back into the games they...
13/06/2025
Industrial AI isn't slowing down. Germany is ready.
Following London Tech Week and GTC Paris at VivaTech, NVIDIA founder and CEO Jensen Huang's Europea...
12/06/2025
In 2018, Spotify launched Heart & Soul, a mental health initiative developed to ...
12/06/2025
50 Years Strong: SBS and NITV Supercharge NAIDOC Week 2025 in a joint 50th celeb...