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...
29/09/2026
Oak View Group (OVG) has announced Cisco as an Official Technology Partner, with plans to standardize infrastructure across its venue network on Cisco solutions...
29/09/2026
SportsNet Pittsburgh (SNP) won three honors at the 2026 Mid-Atlantic Emmy Awards.
Program awards went to Bucs BP in the Sports Program Live category (produced ...
29/09/2026
The Professional Women's Hockey League has announced home openers for all 12 teams for the 2026-27 season, along with a broadcast partnership with CBC under...
29/09/2026
Manifold technologies has been recognized as a contributor to the Media eXchange...
29/09/2026
FuboTV has announced My Local Sports, allowing subscribers in select markets to add their local regional sports network (RSN) as an add-on to a Fubo Sports subs...
29/09/2026
Asian News International (ANI), a wire agency and video news provider based in India, has deployed Grass Valley's AMPP Media Operating System and Framelight...
29/09/2026
South African Broadcasting Corporation (SABC) has deployed Ikegami UHK-X600 cameras as part of an upgrade to IP-based production infrastructure. The deployment ...
29/09/2026
REMI, REMCO models will drive the bulk of the broadcaster's NHL productions ...
29/09/2026
Three years in, the facility is supporting more centralized race production, com...
29/09/2026
NBC, Peacock, and NBSN are carrying all four Wild Card series with a production ...
29/09/2026
19 Nonfiction Projects Across All Stages of Production to Receive More Than $1 million in Support as Documentary Film Program Approaches 25th Anniversary
LOS ...
29/09/2026
Two new plug-ins launch alongside updated bundles
Zynaptiq have just announced the launch of two new plug-ins, alongside new and improved versions of two of...
29/09/2026
All-in-one bus processor revealed
Wes Audio's ng product range has come a long way since it was first introduced, and now includes a tasty selection of ...
29/09/2026
eds3_5_jq(document).ready(function($) { $(#eds_sliderM519).chameleonSlider_2_1({ content_source:......
29/09/2026
Share
Copy link
Facebook
X
Linkedin
Bluesky
Email...
29/09/2026
Share
Copy link
Facebook
X
Linkedin
Bluesky
Email...
29/09/2026
Share
Copy link
Facebook
X
Linkedin
Bluesky
Email...
29/09/2026
Share
Copy link
Facebook
X
Linkedin
Bluesky
Email...
29/09/2026
Ikegami announces that the South African Broadcasting Corporation, SABC, has chosen Ikegami UHK-X600 camera systems as part of an advance to IP-based operation ...
29/09/2026
Blendkit Offers 20% Discount to Celebrate Crossing 150K Asset Milestone
Brie Clayton September 29, 2026
0 Comments
Blendkit, the largest 3D asset libr...
29/09/2026
Interactive streaming specialist Ease Live supported more than 76.5 million fan sessions across three major European media operators during this summers FIFA Wo...
29/09/2026
Ease Live has added support for VAST (Video Ad Serving Template) to its graphical overlay technology, opening up new ways for broadcasters, streamers and rights...
29/09/2026
Rise AV has announced the launch of the inaugural Rise AV Awards, a new global awards programme celebrating the women, advocates and organisations across the AV...
29/09/2026
Share
Copy link
Facebook
X
Linkedin
Bluesky
Email...
29/09/2026
Share
Copy link
Facebook
X
Linkedin
Bluesky
Email...
29/09/2026
Share
Copy link
Facebook
X
Linkedin
Bluesky
Email...
29/09/2026
Share
Copy link
Facebook
X
Linkedin
Bluesky
Email...
29/09/2026
Share
Copy link
Facebook
X
Linkedin
Bluesky
Email...
29/09/2026
Share
Copy link
Facebook
X
Linkedin
Bluesky
Email...
29/09/2026
Clear-Com announced that First Baptist Nashville has upgraded its production communications with a Clear-Com intercom solution as part of a comprehensive sanct...
29/09/2026
Blackmagic Cinema Camera 6K Captured Action Shots for Sketch
Brie Clayton September 28, 2026
0 Comments
Camera's portable design and dynamic range...
29/09/2026
Kean Production Services Deploys AJA ColorBox to Deliver Consistent, Accurate Co...
29/09/2026
Berklee Alumni and Faculty Nominated for the 2026 Latin Grammy Awards A total of 28 Berklee alumni and faculty earned nominations across 20 categories this ye...
29/09/2026
Berklee's K-Pop & Beyond Explores Connections Between K-Pop and Hip-Hop Now in its fifth year, the series brings together artists, industry leaders, and s...
29/09/2026
We tend to think we live in a digital world, with hopes and fears around artificial intelligence (AI), quantum computing, data centres and the next big technolo...
29/09/2026
Today, in partnership with the American Academy of Pediatrics (AAP), Apple relea...
29/09/2026
Final Cut Camera, the professional filming app for iPhone, receives a major upda...
29/09/2026
RT lyric fm is proud to announce the annual Choirs for Christmas competition callout. Warm up those vocal chords and join us on RT lyric fm for a celebration ...
28/09/2026
The NHL has announced Adobe as an Official Creative, Marketing and AI Partner beginning with the 2026-27 season. The partnership will provide the league and its...
28/09/2026
MultiDyne will exhibit at CABSAT 2026 (October 5-7, Dubai Exhibition Centre), showcasing an expanded technology portfolio following its June 2026 acquisition of...
28/09/2026
ARRI and HONOR have announced the HONOR Magic9 Pro Max, a flagship smartphone carrying ARRI branding on the device and packaging as the first product-level expr...
28/09/2026
Net Insight has added the Nimbra 1020G to its pan-Asian live media trial network, giving customers and partners in the region a way to evaluate ST 2110 and JPEG...
28/09/2026
Vindral has been named a recipient of a 2026 Technology and Engineering Emmy Award by the National Academy of Television Arts and Sciences in the category of We...
28/09/2026
Interra Systems will exhibit at CABSAT 2026 (Stand B2-37, Dubai Exhibition Centre, October 5-7), demonstrating its BATON QC platform, ORION monitoring solutions...
28/09/2026
Two years into its new production facility, the TOUR is expanding centralized wo...
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.
