Latest news with #machinelearning
BBC News
11 hours ago
- Climate
- BBC News
Weather forecasts: The tech giants use AI but is it any good?
A wave of machine-learning weather models have been unleashed by some of the very biggest businesses on the challenge the orthodoxy of traditional physics-based computer forecasts that have been incrementally developed and improved over many decades. But are the machine learning models any good? The weather is a national obsession for us Brits, and it is no wonder given the huge changes that are seen and felt from one day to the next. Accurate weather forecasts are not just vital for planning our daily lives but knowing about upcoming severe weather can help us to change our behaviour, save lives and mitigate damage to is impossible to assess the full economic value of weather forecasts globally, but the numbers are huge. According to NOAA (National Oceanic and Atmospheric Administration), in the US alone - and just taking into account the biggest weather disasters that caused over $1bn (£740m) in damage - the fallout from severe weather in 2024 amounted to $182bn, with 568 1980, this damage figure stands at nearly $3tn! Meanwhile, in the UK, there were 1,311 excess deaths caused by heatwaves in 2024.A study from consultants, external, London Economics, concluded that the Met Office would bring £56bn of benefits to the UK economy over the course of a decade through providing meteorological services. Multiply these kinds of numbers across the whole world, with a growing population exposed to increasingly extreme weather fuelled by climate change, and weather is big business. The biggest computers on the planet Traditional weather forecasts are produced on some of the biggest supercomputers on the planet; the Met Office super computing contract is worth £1.2bn. That huge sum of money buys you a machine that can perform 60 quadrillion (60,000,000,000,000,000) calculations per second, running a model containing the understood physics, with over a million lines of code and using 215 billion weather observations. Global weather models work by crunching the numbers in a grid of boxes right around the planet. The size or resolution of these boxes varies across different meteorological models, but range between about 10sq km to 28sq km (3.86sq miles to 10.81sq miles).At this kind of resolution they cannot accurately predict showers, while mountain ranges are lower and smoother than in the real world. The highest resolution model from the Met Office, the UKV (the model that runs BBC TV graphics for the first 48 hours) can predict showers with its incredible 1.5km (0.9 mile) resolution - but it takes so much computing time and power that this model is not able to forecast for the entire world; instead it concentrates on the UK and Europe. Machine-learning weather models: are they any good? Machine-learning weather models have only been around for a few years, showing promise as they develop rapidly. Traditional models take hours to run on hugely expensive supercomputers, however this new breed of models can take less than a minute to run on a standard laptop. They don't need to know all the 'burdensome' laws of physics, but are instead trained on 40 years of past data to make their do they perform? Well let's look at forecast verification data from ECMWF (European Centre for Medium Range weather Forecasting) for atmospheric pressure patterns in winter 2024/2025, (Google), AIFS (ECMWF) and Aurora (Microsoft) were more accurate than the traditional IFS (ECMWF) benchmark forecast, whereas FourCastNet (Nvidia) and Pangu-Weather (Huawei) trailed some of the machine-learning models performed better, and some worse, but it depends on which variable you look at, and all of this could change quickly as the rate of progress accelerates. Just like traditional models, AI models are less accurate the further ahead in the future they're trying to predict - a consequence of the chaotic nature of the atmosphere. Looking 10 days ahead, none of the AI models (or traditional models) were able to offer forecasts considered to be of much use in terms of accuracy. So is it time to walk away from physics-based weather models? Not yet! Machine-learning weather models are not only trained using data produced by traditional weather models, but they also use the start position of the atmosphere from traditional models as their input point. In other words, without those traditional models running, the machine-learning models wouldn't work as machine-learning models can forecast large scale features like high and low pressure six days ahead very well, but they can underperform compared to traditional models at smaller-scales of 1000km or means that important features like troughs and ridges could be missed, which would make the difference between a dry day, or a day with heavy rain. The majority of machine learning models have a resolution of 28sq km, which is the same scale as the data that they've trained on. This means small features like showers would most likely be missed, so they wouldn't be able to forecast a Boscastle flood event many days headlines have claimed that these new models are better than traditional models at predicting hurricanes. It may be true that some have been a little better at predicting the landfall of hurricanes ahead of traditional models, but at the same time they have been very poor at predicting the wind strength and therefore the likely damage the storm would bring. This may be the result of the smoothing, or averaging effect of looking at lots of hurricanes in the 40 years of training data. AI models may struggle to forecast effects from rare events that have not been seen often in the 40 years of training data. An example of this is the 1991 eruption of Mount Pinatubo that cooled the planet down by up to 0.5C for two are also questions about how well AI models will forecast in a warmer world as our planet continues to heat up as a result of climate change. The past climate that they've trained on will look quite different to our future climate, as greenhouse gases continue to accumulate in the where will we be in five years time?"I think we'll have traditional models running alongside AI models so that we are drawing on their combined strengths to enable hyper-localised accurate forecasts, delivered fast, when you need them," says Professor Kirstine Dale, chief AI officer at the Met Office. Machine-learning models haven't been around for very long, but with their speed, computer efficiency and rapid rate of development, they show great potential.
Android Authority
3 days ago
- Business
- Android Authority
Google's updated Gemini 2.5 Flash pricing is good news and bad news
TL;DR Gemini 2.5 Flash-Lite is rolling out in preview. The stable versions of Gemini 2.5 Flash and Pro are also rolling out. The cost of input for Flash has gone up by $0.15, but the output has been reduced by $1.00. Google introduced Gemini 2.5 earlier this year, with Pro (experimental) being the first model in the series. A couple of months later, the company rolled out a faster model in early access, known as Gemini 2.5 Flash. Now the firm has an update on these two models and is launching an even faster model in preview. It may be hard to keep track of all the different Gemini models at this point, but Google is debuting a third model in the 2.5 line. This model is called Gemini 2.5 Flash-Lite and is said to have the lowest latency and cost in the family. According to Google, this model is best for 'high throughput tasks like classification or summarization at scale.' Like Gemini 2.5 Pro and Flash, Flash-Lite is described as a reasoning model. This means it's capable of reasoning through its thoughts before responding for better accuracy. Unlike the other two models, Google says 'thinking' is turned off by default on Flash-Lite since cost and speed are the focus. However, the model 'allows for dynamic control of the thinking budget with an API parameter.' Meanwhile, the Mountain View-based firm is now making Gemini 2.5 Pro and Flash stable and generally available. Along with going stable, Google says it's updating the pricing for Flash. The cost for input is going up by $0.15, while the output cost has been reduced from $3.50 to $2.50. You can continue using the existing pricing scheme for 2.5 Flash Preview or 2.5 Pro Preview until the depreciation date. For 2.5 Pro Preview, the depreciation date will be June 19, 2025. You'll have until July 15, 2025, before Google shuts down 2.5 Flash Preview. Got a tip? Talk to us! Email our staff at Email our staff at news@ . You can stay anonymous or get credit for the info, it's your choice.
Forbes
3 days ago
- Business
- Forbes
AppLovin Stock: Worth It At $365?
The AppLovin Corporation logo appears on a smartphone screen in this illustration photo in Reno, ... More United States, on December 20, 2024. (Photo by Jaque Silva/NurPhoto via Getty Images) AppLovin (NASDAQ:APP), a firm that assists mobile app developers in publishing and promoting their applications, has excelled over the past year, though it has experienced some ups and downs in 2025. While the stock dropped nearly 57% from the peaks observed in early February 2025 following a report from a short-seller claiming AppLovin violated service terms and exaggerated the success of its e-commerce operations, these claims have yet to be substantiated, and the stock saw a strong recovery after reporting solid earnings in the first quarter. The stock is currently approximately 7% up year-to-date in 2025 and has increased nearly 4.5x over the previous year. Demand for Axon 2.0, AppLovin's exclusive machine learning algorithm for ad placement, has surged. The software effectively determines which ad to show, to which user, and at what time in order to optimize click-through rates or user engagement. While this resembles the strategies employed by Meta and Google, Axon is specifically designed for mobile app advertising. The company's advertising platform reported impressive revenue growth of 71% year-over-year in Q1 2025, achieving $1.16 billion. Overall financial results have also been strong, with revenue soaring nearly 40% year-over-year, and adjusted EBITDA experiencing an increase close to 83%. Although the bulk of the company's revenue still comes from advertisements for mobile gaming applications, it is concentrating on expanding its e-commerce sector. Nevertheless, it remains uncertain how effective this initiative will be, as AppLovin possesses an extensive dataset in gaming but may lack the comprehensive first-party e-commerce data that competitors like Meta and Alphabet have. That being said, despite its impressive growth and success, the AppLovin stock may be a challenging choice at its current price of approximately $360. We believe there is little reason for concern regarding APP stock, which makes it appealing, but it is also very responsive to negative events due to its elevated valuation. We reached this conclusion by comparing APP's current valuation with its operational performance over the past few years, as well as its current and historical financial positions. Our evaluation of AppLovin according to key metrics such as Growth, Profitability, Financial Stability, and Downturn Resilience indicates that the company possesses a very strong operational performance and financial standing, as elaborated below. Nonetheless, if you are looking for potential with lower volatility than individual stocks, the Trefis High Quality portfolio offers a viable alternative - having outperformed the S&P 500 and delivered returns greater than 91% since its founding. Separately, see – Should You Buy CRWV Stock After A Whopping 4x Rise? Based on the amount you pay per dollar of sales or profit, APP stock appears to be very expensive relative to the overall market. • AppLovin has a price-to-sales (P/S) ratio of 25.1 compared to a figure of 3.0 for the S&P 500 • Additionally, the company's price-to-free cash flow (P/FCF) ratio stands at 50.8 compared to 20.5 for the S&P 500 • Furthermore, it has a price-to-earnings (P/E) ratio of 67.1 compared to the benchmark's 26.4 AppLovin's Revenues have substantially increased over the last few years. • AppLovin has experienced an average revenue growth rate of 23.2% over the last 3 years (versus an increase of 5.5% for the S&P 500) • Its revenues have risen 41.6% from $3.6 billion to $5.1 billion in the past 12 months (compared to a rise of 5.5% for the S&P 500) • Moreover, its quarterly revenues increased by 40.3% to $1.5 billion in the most recent quarter from $1.1 billion a year prior (versus a 4.8% increase for the S&P 500) AppLovin's profit margins are significantly higher than most companies within the Trefis coverage universe. • AppLovin's Operating Income over the last four quarters was $2.4 billion, indicating a significantly high Operating Margin of 46.5% (compared to 13.2% for the S&P 500) • AppLovin's Operating Cash Flow (OCF) for this period was $2.5 billion, indicating a significantly high OCF Margin of 49.4% (versus 14.9% for the S&P 500) • For the last four-quarter span, AppLovin's Net Income was $1.9 billion - reflecting a significantly high Net Income Margin of 37.4% (compared to 11.6% for the S&P 500) AppLovin's balance sheet appears solid. • AppLovin reported a Debt figure of $3.7 billion at the end of the most recent quarter, while its market capitalization is $124 billion (as of 6/13/2025). This results in a very strong Debt-to-Equity Ratio of 2.9%(in contrast to 19.9% for the S&P 500). [Note: A low Debt-to-Equity Ratio is preferable] • Cash (including cash equivalents) constitutes $551 million of the $5.7 billion in Total Assets for AppLovin. This gives rise to a moderate Cash-to-Assets Ratio of 9.7% (in comparison to 13.8% for the S&P 500) APP stock has underperformed compared to the benchmark S&P 500 index during several recent downturns. While investors remain hopeful for a smooth economic adjustment in the U.S., the potential consequences of another recession could be severe. Our dashboard How Low Can Stocks Go During A Market Crash illustrates the performance of key stocks during and after the last six market crashes. • APP stock declined 91.9% from a high of $114.85 on 11 November 2021 to $9.30 on 27 December 2022, compared to a peak-to-trough drop of 25.4% for the S&P 500 • The stock completely rebounded to its pre-Crisis peak by 16 September 2024 • Since then, the stock has risen to a maximum of $510.13 on 17 February 2025 and currently trades at around $360 • APP stock dropped 36.7% from a high of $88.22 on 17 June 2021 to $55.88 on 16 August 2021, in comparison to a peak-to-trough decline of 33.9% for the S&P 500 • The stock completely recovered to its pre-Crisis high by 14 October 2021 In conclusion, AppLovin's performance across the parameters mentioned previously can be summarized as follows: • Growth: Extremely Strong • Profitability: Extremely Strong • Financial Stability: Very Strong • Downturn Resilience: Weak • Overall: Very Strong Therefore, despite its high valuation, the stock seems attractive yet volatile due to its poor downturn resilience. This leads us to affirm that APP is a challenging stock to purchase. Not satisfied with the volatility associated with APP stock? The Trefis High Quality (HQ) Portfolio, which consists of 30 stocks, has demonstrated a history of comfortably outperforming the S&P 500 over the past four years. Why is this the case? As a collective, HQ Portfolio stocks have provided superior returns with less risk compared to the benchmark index; offering a smoother experience, as shown in the HQ Portfolio performance metrics.
South China Morning Post
3 days ago
- Business
- South China Morning Post
Alibaba updates open-source Qwen3 models to support AI deployment on Apple devices
Advertisement In a Monday post on social media platform X , the Qwen team of Alibaba's cloud computing unit said it launched open-source Qwen3 models optimised for Apple's MLX framework for machine learning. Alibaba owns the South China Morning Post. The MLX framework is intended to be user-friendly, but still efficient in training and deploying AI models on Apple's silicon hardware. It has seen growing adoption among developers within the Apple ecosystem. Open source gives public access to a software program's source code, allowing third-party developers to modify or share its design, fix broken links or scale up its capabilities. This latest development appears to reflect Apple's efforts to expand the availability of its Apple Intelligence suite to mainland China. Under Chinese rules, any models that are part of Apple Intelligence would need regulatory approval for public release in the country. Advertisement
Bloomberg
4 days ago
- Business
- Bloomberg
Robot Maker 1X Launches Model to Train Humanoids Faster
Robotics firm 1X Technologies launches a new "World Model," which it says is the first data-driven simulator for humanoid robots allowing the machines to learn to move in the physical world faster. 1x CEO and CTO Bernt Børnich speaks with Caroline Hyde and Ed Ludlow on 'Bloomberg Tech.' (Source: Bloomberg)
