
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 ...
09/10/2026
September 10 2026, 06:00 (PDT) Dolby Expands Dolby OptiView Platform with New Capabilities at IBC 2026
New Sports Intelligence helps providers better unders...
07/10/2026
Dalet, a leading technology and service provider for media-rich organizations, today announced the latest Long-Term Supported (LTS) release of Dalet Flex. Build...
11/09/2026
Calrec will introduce two new products at IBC2026 (Stand 8.C47, RAI Amsterdam, September 11-14): the Type RT broadcast production console and the Apex M6 Modula...
11/09/2026
Grass Valley has announced AVORA, a standalone all-in-one live production solution launching at IBC2026 (Booth 9.A01 and IBC Future Tech Zone, Stand 14.C49, Sep...
11/09/2026
CAMB.AI is introducing its Streaming SDK and Streaming Dashboard at IBC2026, wit...
11/09/2026
EVS has announced XT-S, a software-defined server platform that runs multiple live production applications on a common hardware foundation. The platform debuts ...
11/09/2026
EKT is demonstrating a new Linux-based ATSC 3.0 set-top box at IBC2026. The second-generation product moves away from Android to reduce memory costs, with Linux...
11/09/2026
The Sennheiser Group will exhibit at IBC2026 (Hall 8, Booth D50, September 11-14...
11/09/2026
ESPN's Monday Night Football will open its 2026-27 season with a new renditi...
11/09/2026
Dalet has announced an integration of NVIDIA's Synthetic Video Detector AI model into Dalet Flex, enabling media organizations to identify potentially AI-ge...
11/09/2026
The Media Coding Industry Forum (MC-IF) will demonstrate Versatile Video Coding (VVC) capabilities at IBC2026 (Hall 2, Stand 2.B13), as the codec moves toward c...
11/09/2026
Sony will exhibit at IBC2026 (Hall 13, September 11-14), demonstrating products across live production, news, cinema, and immersive and spatial content workflow...
11/09/2026
Wave Central, a Domo Broadcast Company, has been named a recipient of a 2026 Tec...
11/09/2026
Studio Network Solutions (SNS) is bringing a pair of workflow updates to IBC2026...
11/09/2026
iodyne will co-exhibit with Hedge at IBC2026 (Booth 7.A36, September 11-14) and appear across three additional partner booths, marking the first public demonstr...
11/09/2026
In-venue and creative video staffers at the professional and collegiate level ha...
11/09/2026
The network will phase in HDR crew by crew while expanding Next Gen Stats integration, SkyDry availability, and onsite studio shows...
11/09/2026
CAMB.AI is introducing its new Streaming SDK and Streaming Dashboard at IBC2026,...
11/09/2026
Massive host operation at Melbourne Cricket Ground featured 54 cameras, 25,000 ST 2110 flows, REMI studio production in Stamford, and global distribution to mor...
11/09/2026
Production Anywhere Remote Kit (PARK) & AR-fueled Stage B are two key pieces to ...
11/09/2026
By Ramona Flume and Lucy Spicer
There's a particular kind of magic that hap...
11/09/2026
By Jessica Herndon
The fall film season is kicking off, and Sundance Institute-...
11/09/2026
Spark series gains compact modelling stompbox
Positive Grid's Spark amps have become a popular part of the company's line-up, offering a convenient ...
11/09/2026
New flagship sequencing plug-in revealed
Rapid Flow have just launched a new 16-track sequencer plug-in which they say has been designed for total rhythmic...
11/09/2026
Rohde & Schwarz unveils the air-cooled R&S TM1 medium-power UHF transmitter at I...
11/09/2026
FIFA World Cup 2026 and Peacock's Love Island USA' Power NBCU-Versant ...
11/09/2026
Share
Copy link
Facebook
X
Linkedin
Bluesky
Email...
11/09/2026
Share
Copy link
Facebook
X
Linkedin
Bluesky
Email...
11/09/2026
Lightware has announced the launch of LARA (Lightware Advanced Room Automation) 2.0, a major upgrade to its built-in room automation platform, introducing impro...
11/09/2026
10 September 2026 - MusiMap, the B2B music intelligence company that is transforming how companies can harness the emotional power of music to reach consumers a...
11/09/2026
Matrox Video today announced an integration between Matrox ORIGIN and NVIDIA Holoscan for Media, enabling Matrox ORIGIN media services to be provisioned, config...
11/09/2026
Net Insight today announced that its cloud-native live media orchestration is now available through Oracle Cloud Marketplace, providing broadcasters, production...
11/09/2026
Share
Copy link
Facebook
X
Linkedin
Bluesky
Email...
11/09/2026
Version 1.8 will bring advanced routing control, new workflow tools, boosted control panel flexibility and ease of integration into IP networks...
11/09/2026
Ikegami announces four additions its UNICAM XE range of 4K-UHD cameras with the introduction of the UHK-X700VN, UHK-X750VN, UHK-X600VN and UHK-X650VN. Developed...
11/09/2026
DHD.audio has chosen IBC2026 as the launch venue for a major new addition to its broadcast sound production platform: the MX3 console range. MX3 retains the mod...
11/09/2026
Share
Copy link
Facebook
X
Linkedin
Bluesky
Email...
11/09/2026
Share
Copy link
Facebook
X
Linkedin
Bluesky
Email...
11/09/2026
Share
Copy link
Facebook
X
Linkedin
Bluesky
Email...
11/09/2026
Share
Copy link
Facebook
X
Linkedin
Bluesky
Email...
11/09/2026
Share
Copy link
Facebook
X
Linkedin
Bluesky
Email...
11/09/2026
Share
Copy link
Facebook
X
Linkedin
Bluesky
Email...
11/09/2026
Share
Copy link
Facebook
X
Linkedin
Bluesky
Email...
11/09/2026
Share
Copy link
Facebook
X
Linkedin
Bluesky
Email...
11/09/2026
New standalone solution gives production teams everything they need to create, produce and deliver live content from one flexible system.
Grass Valley today an...
11/09/2026
Recognition honors the development of FPV drone camera systems for live sports broadcast coverage
Wave Central, a Domo Broadcast Company that develops advanced...
11/09/2026
Teradek, the industry leader in zero-delay wireless video solutions, announced the launch of Bolt 7, which now includes a unified wireless video and data system...
11/09/2026
The ACT Entertainment tvONE and Green Hippo Pixel Academy has released its autumn course schedule, with hands-on training dates in September, October and Novemb...
11/09/2026
LA JOLLA, CA-Transthyretin amyloid disease (ATTR) is a progressive, life-threatening condition caused when the protein transthyretin (TTR) breaks apart and aggr...