Articles

Former des données avec ChatGPT : Guide pour développeurs

Apprenez à former des données avec ChatGPT et découvrez comment les développeurs peuvent tirer le meilleur parti de cette technologie puissante !

## Le lancement de ChatGPT par OpenAI a été transformateur pour l’intelligence conversationnelle AI. Impressionnant hors de la boîte, les capacités de ChatGPT sont intrinsèquement limitées par ses données d’entraînement fixes de 2021. Pour les développeurs de logiciels et les entreprises technologiques, l’entraînement de ChatGPT sur des jeux de données personnalisés est essentiel pour créer des assistants IA personnalisés qui évoluent avec votre entreprise.

Dans ce guide complet, nous explorerons les meilleures pratiques pour les équipes de logiciels afin de former des modèles ChatGPT personnalisés à l’aide de techniques telles que le réglage fin et la lecture interactive MEMWALKER.

Testing is a critical part of training ChatGPT models. It’s important to evaluate the performance of your model against a test dataset to ensure that it’s accurately predicting the desired output. Testing also helps identify any potential issues with the model, such as overfitting or underfitting. To get the most out of testing, it’s important to use a variety of metrics, such as accuracy, precision, recall, and F1 score.

Le lancement de ChatGPT par OpenAI a été transformateur pour l’intelligence conversationnelle IA. Impressionnant à l’état brut, les capacités de ChatGPT sont intrinsèquement limitées par ses données d’entraînement fixes de 2021. Pour les développeurs de logiciels et les entreprises technologiques, l’entraînement de ChatGPT sur des ensembles de données personnalisés est essentiel pour créer des assistants IA adaptés à votre entreprise.

Dans ce guide complet, nous explorerons les meilleures pratiques pour les équipes de logiciels afin de former des modèles ChatGPT personnalisés à l’aide de techniques telles que le réglage fin et la lecture interactive MEMWALKER.

Le test est une étape essentielle de l’entraînement des modèles ChatGPT. Il est important d’évaluer les performances de votre modèle sur un jeu de données de test pour s’assurer qu’il prédit correctement la sortie souhaitée. Les tests permettent également d’identifier tout problème potentiel avec le modèle, tel que le surapprentissage ou le sous-apprentissage. Pour tirer le meilleur parti des tests, il est important d’utiliser une variété de métriques, telles que la précision, la précision, le rappel et le score F1.

Source de l’article sur DZONE

Designing for user experiences is what all designers do. UX is often thought of as the preserve of app or web designers; however, even a print designer laying out a magazine anticipates reader reaction to the scale of type, the placement of adverts, and the art direction of successive stories.

Because all designers design user experiences, the role of UX Designer has come to mean someone focused on creating a product or service utilizing research and testing to guide decision-making.

To research and test anything, you need metrics: a baseline and a target against which to measure. No one set of metrics is suitable for all projects, but because UX tends to be for financial profit, the Pirate Metrics Framework — Acquisition, Activation, Retention, Referral, Revenue — is a good starting place.

You might seek out very different metrics in some cases. For instance, a museum might measure the success of its education program based on how many students go on to study paleontology. However, those types of metrics are notoriously difficult to quantify. Excepting a few niche cases, successful UX increases user productivity, decreases errors, reduces the cost of support, and increases sales.

So if it’s as easy as counting dollars, why does UX go bad?

UX vs. Design Principles

To understand what UX is, you need to understand what UX is not.

One of the most straightforward design principles to understand is hierarchy: bigger is more important, i.e., a heading is visually stronger than a sub-heading, a sub-heading is visually stronger than the body text.

Design principles stem from one thing: human-centered design. At the most basic level, bigger is more important because the bigger a saber-toothed tiger appears, the more likely it intends to eat me.

The evolution of human beings is so slow that had a smartphone existed at the time, a neanderthal would have been able to tap a button with the same level of precision as me. Prehistoric man shares the same minimum button size as modern man: 48 x 48px. Design principles don’t change, don’t require research, and don’t need verifying with tests.

On the other hand, a neanderthal would not have understood a smartphone, let alone an app. You only need to step back by a single generation to find perfectly intelligent people baffled by a commonly employed design pattern.

Unlike design principles, user experience is a house built on sand. When the sand shifts, the walls crack. The bricks are still solid, but the rain gets in.

Because effective UX is temporary, so is the ROI.

Technology Breaks UX

Technology unfolds at a rapid pace. As technology develops, the user experience defined by that technology changes.

The classic example is the mobile revolution, but technological change does not necessarily mean hardware. One of the most significant shifts in UXD (User Experience Design) in my career has been the popularisation of AJAX — the process of using JavaScript to load new data without refreshing the page. This seamlessness has been around since the early 2000s, but it’s only in the last ten years, as the code to achieve it has simplified, that it’s been widely used.

Jakob’s Law states that users spend most of their time on other sites and, as a result, prefer your site to function like other sites by following familiar design patterns.

Even if your UX is rigorously tested and optimized, when other sites and services carry out their own research, they are testing against the background of younger technology, and the “other sites” Jakob Nielsen refers to begin to change. As a result, the UX of your site is gradually eroded.

The consequence of continual technological change is that user research is constantly invalidated. The UX of an app, site, or service begins to degrade as soon as it is created.

User-Experience Lifecycle

Human beings have two deep-seated motivations: survival and procreation. The most important, survival, depends on discovery — new food sources, new routes through dangerous territory, new ways to skin a mammoth. We are biologically programmed to seek out the new.

A typical user passes through three phases of a relationship with a site, app, or service: discovery > comfort > boredom. Churn, or drop-off, tends to occur in the discovery phase (if the comfort phase is too slow in developing) or the boredom phase. The sweet spot is the comfort phase. That’s the part of the business-customer relationship in which the customer requires minimal support and is least likely to drop off.

The most effective form of UX — meaning the one that satisfies most metrics — rapidly moves a user from discovery to comfort and then continually eases the user back to the start of the comfort phase without tipping back into discovery.

This can be achieved with numerous micro-discoveries, tiny chunks of new experience, from simple functionality tweaks to style revisions.

Summary

All UXD, regardless of the quality, level of investment, and skill of the practitioner, begins to degrade the moment it is created.

Design principles like simplicity are good indicators of successful UID (User Interface Design) and are timeless; comprehensive design systems, brand assets, and content offer good ROI.

The most effective UX is broadly familiar and continually refreshed in small ways, allowing users to enjoy the comfort of the familiar while also experiencing the excitement of discovery again and again.

 

Featured image uses photos by Wolfgang Hasselmann & Shainee Fernando.

Source

The post When UX Goes Bad (and How to Fix It) first appeared on Webdesigner Depot.

Source de l’article sur Webdesignerdepot

Maps are a fascinating method for delivering content. At their best, they can create an intuitive way of presenting information and interacting with it. This is the advantage that digital maps, through mobile apps and websites, have over print maps and images where no interactivity is possible.

But it’s important to understand that more data ≠ better experiences. We all now have so much data available to us through multiple services that, arguably, the greatest challenge isn’t sourcing information but filtering it out. We can only handle so much information input before we become overloaded. This issue risks being omnipresent with maps. There are so many potential points of interest on a map that it’s essential to be clear about what needs to be exposed to users.

Also, UX design, map design, and user interface are all critical. While maps can be a powerful way of drawing people in, if end-users feel that you didn’t even consider the visual design, they’ll ‘bounce off’ your site or app in moments.

Common Use Cases

When are maps useful, and what problems do they solve? Let’s dive right into the most common use cases for maps used in web design.

Navigation and Direction

Like Google Maps shows, navigation and direction are arguably the classic case study for interactive maps. You are in one place and need to get to another. You can enter your destination, your current location, and the map will present suggestions for getting there. You can select the method of travel and adjust desired departure or arrival times. But you need to understand first what functionality your users need. How these options are exposed to users is a critical piece of UX design.

Also, if users are searching for options such as somewhere to eat, it’s not so straightforward. Then, how your map handles panning in real-time as users swipe around a city is going to be a big issue.

Showing Relationships and Trends Geographically

This is something that you’ll see in every election in any western country. We’re all used to seeing maps that give us a state-of-play for which state or seat is held by which party. Then, we might see projections based on voter intentions and projected voting swings deriving from that. Then, exit poll data can be projected with the map updated on an ongoing basis until the final result is confirmed.

The capability to do this is essential because if a static map were used, it’d be out of date any time a new poll was released. Also, voting intentions can change over a campaign, so such maps need to be dynamic. Of course, such maps are only as accurate as the available data, as the US 2016 election map showed.

Show Points of Interest

As mentioned previously, there’s a lot of data that can be exposed to map users. However, that doesn’t automatically mean that it should be. Usability is key. For example, when you look at a map, you’ll typically first see key points of interest. Which points of interest are going to be presented to you can vary.

One variant is zoom level. If your map is currently showing an entire city, the level of detail the map presents is deliberately limited. You’ll see districts, large roads, or geographic features such as rivers. If more detailed information were presented, users on mobile devices, in particular, would be overwhelmed. Even at this level, you’ll notice typography differences. These can include the city name being in bold or the names of different areas in capital letters. So the level of detail is coupled with the scale of the map. Zooming in a few notches will expose significant points of interest, such as museums. Zooming in to specific districts will reveal restaurants, coffee shops, and universities. This visual hierarchy is a critical way of managing the exposed level of information.

But information is still being abstracted away. It’s not until you tap on the museum that you’ll see information on opening hours and busy times. This is also typically presented with user photos and reviews. Context is also taken into account, so you’ll start to see local hotels and restaurants. So it’s not just individual points of interest that are important, but the connections between them.

6 Tips For Improving Interactive Maps

What are the challenges of creating effective maps, and how do people address the data overload problem? We’ll answer this question and go over the must-know aspects of map creation.

1. Ensure Security and Brand Trust

GDPR or General Data Protection Regulation. This is a critically important European law that extends a wide range of legal protection to European citizens regarding personal data. It’s not possible here to cover the full extent of the law, but here are some quick key points:

  • Consent is required for the processing of personal data; it cannot be assumed
  • You need to have a retention policy for information that’s capable of identifying people

Be aware that the latter doesn’t just cover commercial purposes. Research students have to submit GDPR forms that address what kind of data they’re sourcing and how they’ll be retaining it.

But the most crucial context is commercial. If a business suffers a data breach, it can be fined up to 20 million euros or 4% of annual worldwide turnover in the preceding financial year, whichever is greater. Therefore, any business storing data that could identify their customers will need to assess risk and compliance. Remember: it’s 4% of worldwide turnover, not EU turnover.

Also, anything of your business that you expose to your customers or users is an extension of your brand. Therefore, you need to assess your maps for brand compliance too. If you have primary brand colors and your map doesn’t abide by them, that’s a very poor look. Source the color hex codes directly from your brand team and involve them in design.

2. Use the Appropriate Type of Map

It’s also important to consider what type of map is most appropriate for your use case. Think carefully about what your users need, what you’re trying to communicate, what information you need to present, and how best to present it.

For example, points of interest style maps in a tourist app will be way more helpful than heat maps: people want to know where something is, key data like opening hours, and how to get there. A heat map showing the number of visitors to each attraction or area of a city is unlikely to be useful to tourists. However, it could be useful to the attractions themselves to map their visitors by heat map over time. This could help larger museums chart which exhibits are most popular.

Transport for London is charting passenger movement on the London Underground by detecting when a device with Wi-Fi comes into range and then passes out of range. They’re using this to understand overall user journeys and movements within individual stations to better manage disruptions.

3. Avoid Pop-Ups

It should go without saying by now that auto pop-ups are despised. It doesn’t matter what they’re doing or what they’re offering; an unwanted pop-up can only get in the way. The level of impact is even greater on a phone where pop-ups take up even more screen space.

Given this, many users close them without even reading them. So if you’re using pop-ups, don’t kid yourself. You’re likely just irritating users and increasing the likelihood that they’ll ‘bounce off’ or uninstall.

4. Avoid Auto-Geolocation

Auto-geolocation sounds incredibly convenient but can result in some real problems. For example, if there are any bugs with auto-geolocation, you could get false results. If someone connects through public building Wi-Fi, you could get false results. If they’re connecting through a VPN then, unless you get the user’s IP address and check if it’s the exit portal of a VPN, you could get false results.

The problem is most significant with mobile maps. If a map user is looking at a points of interest map, they likely have a specific and immediate use. This means it’s in their best to get the most accurate results possible. So why not just ask them?

Precision and Accuracy

These terms have specific meanings in geolocation. ‘Precision’ is the exactness of the data. ‘Accuracy’ is how closely the information on a map matches the real world. So you want precision and accuracy to be spot on, or data risks losing value. This applies not just to the gathering of data but to the representation of it. For example, if you have street-level data but your maps don’t present individual streets, then any representation of data on that map is likely to have poor accuracy. That map might succeed in abstracting irrelevant information but presenting an imprecise and inaccurate view.

5. Avoid Map Legends as Much as Possible

In many cases, primarily points-of-interest maps, they’re just not needed anymore. An essential part of user experience design isn’t just visual hierarchy but information hierarchy. You can mouse over on a desktop or laptop to get the essentials of a location, e.g., the museum’s name and its opening hours. On a mobile device, you can tap on that location to get the essentials, and you can tap on another location to move on; you don’t even have to press back. Given that, a legend would get in the way. So this simple piece of information design solves information overload issues.

As with all rules, there are exceptions. A good one is a heat map where a density of what’s being measured needs to be communicated. It doesn’t matter what the data is; it just needs to be something where mapping provides greater insight, especially if it informs decision-making. Sales is an excellent example for a national or multinational company. Of course, weather forecasting can make use of literal heat maps.

6. Accessibility Compliance

Not everyone has perfect eyesight. Even if someone has excellent vision, they could still be colorblind (8% of men and 0.5% of women are). Given that, take the W3C’s accessibility standards into account and treat them as a baseline or minimum barrier to entry for compliance. You shouldn’t feel good about the possibility of excluding 8% of your potential audience or customers. Ensure you keep your UX designers involved and don’t shy away from creating senior-friendly web designs.

Put simply: imagine if you could appeal to a new demographic that’s not catered to. If your competitors ignore them, you could give them a real reason to choose you instead by taking some straightforward steps. If your competitors are catering to them, you also need to. If you don’t, you’re just giving potential customers a big reason to ignore you.

Conclusions

The key takeaway is that there’s far more to creating good maps than just good cartography. That can be critical, too, though this may vary depending on the use case.

This will be a team effort because your map will involve data sets, design decisions, and, yes, cartography. You’re going to need to involve brand and IT too. So think about design principles and development methodologies.

First and foremost, what are your users’ needs? If you haven’t done any user research or taken the time to understand the customer journey, are you adding anything or getting in the way? It’s easy to see the department that requested the map as stakeholders, but you should probably view your users as stakeholders too.

This sounds complex, but as you hopefully now appreciate, a map is probably more complicated than you thought.

 

Featured image via Pexels.

Source

The post The UX of Maps in Web Design first appeared on Webdesigner Depot.

Source de l’article sur Webdesignerdepot

Illustration is a big trend in 2021. Every designer should have some ability to illustrate, whether that’s producing icons, creating lettering, mocking up layouts, or crafting full-blown illustrations.

Designers’ workflows are as varied as their work, and one designer’s favorite tool is a frustrating mess to the next designer. Most design apps have a free trial, so it’s worth trying out a few before committing.

Counting down to the best illustration app for designers in 2021, all of the apps on this list are worthy of at least a second glance:

10. Vectr

Vectr is a simple vector design tool. It is heavily geared towards layout and even product design. It’s difficult to use it for many illustration jobs, but simple icon and UI creation are possible. It scrapes into our top ten thanks to the fact that it’s free to use.

9. Paper

Paper is a sketching app developed by WeTransfer. It grabs number nine in our top ten thanks to its simplified interface that is perfect for generating ideas. It may not be the most flexible app, but it’s free to download, making illustration accessible to anyone with an iOS device.

8. Inkscape

We love the fact that there’s someone out there producing a professional-grade artwork app for free. Inkscape is available for GNU/Linux, Windows, and macOS. If you’re interested in exploring this design area without signing up for a subscription or buying an iPad, Inkscape is a great place to start.

7. Assembly

Assembly is another app that’s dependent on the transition of illustrators from desktop machines to mobile devices and is available for iOS. Assembly does things a little differently; instead of manipulating boolean curves, Assembly builds illustrations out of shapes. If you’re someone who thinks they can’t draw (spoiler alert: everyone can draw), this is the app for you.

6. Sketch

It’s a shock low-rating for Sketch, the highly popular product design app doesn’t make it out of the bottom half of our list. The reason for the lower ranking? Sketch is an incredible design app, but its greater investment in prototyping features has come at the cost of drawing innovation. It’s still a great choice for crisp icon design, but it’s not flexible enough to take a higher spot in our list.

5. CorelDRAW

If you’re getting the impression that design apps are mostly for macOS or iOS, then you’d not be alone. CorelDRAW Graphics Suite is one of the few that began its life on Windows before being rebuilt for Mac. CorelDRAW is a premium option, but like Adobe’s offering, it is a suite of tools that cover everything you’ll need to create any form of design work you could ever need.

4. Vectornator

Moving closer to the top spot, Vectornator is an up-and-coming tool. Like many other apps further up our chart, Vectornator does a lot more than simply illustration. Its underlying tools provide a solid foundation for a flexible set of tools that you can use for any artwork. Vectornator does an awful lot, awfully well. 

3. Procreate

Procreate is one of the finest apps on this list, thanks in main to its incredible brush studio; being able to edit the marks you make is essential for any expressive artwork. Procreate enables expressive illustration, typography, and simple sketches. The only downside to this app is that it’s iOS only; if available on more platforms, it might have been even further up the list.

2. Adobe Illustrator

That’s right; we don’t think design giant Adobe’s illustration app Illustrator is worthy of the number one spot.

Illustrator is an excellent tool and arguably deserves to be considered Adobe’s flagship product. Since Adobe relented and allowed individual app subscriptions, it is also reasonably priced. It does everything you could want in an illustration app, from designing icons to creating original artwork.

So why not the top spot? It wasn’t so much that Illustrator fell down anywhere, as a different app pipped it to the post…

1. Affinity Designer

The best app for illustration in 2021 is Affinity Designer. We’ve been fans since its first beta version, and we’ve watched it go from underdog to champion.

On top of being an excellent vector app, Designer is comfortable with raster graphics. Although it certainly can’t compete with Photoshop (Affinity built Photo for that), Designer is more than adequate for many image tasks, making it a great all-around design app. Designer was also the first design app to support macOS’ new M1 chip natively.

The knockout blow is that Affinity Designer is available as a desktop app and an iOS app; the two apps have minor differences in their interfaces; what matters is that both apps use the exact same file format. You can seamlessly switch between desktop, iPad, and back to desktop, taking advantage of the precision of a mouse and the expression of a stylus.

It’s this kind of flexibility that bends Affinity Designer to your workflow instead of the other way around, and that’s why it’s top of our list.

Source

The post Ranked: Top 10 Illustration Apps in 2021 first appeared on Webdesigner Depot.


Source de l’article sur Webdesignerdepot

Some of the changes we are seeing with where we work are starting to pop up in the type of new tools made for designers and developers. More tools with remote collaboration as a key feature are increasing in popularity. (You’ll find a few of those here.)

Here’s what new for designers this month.

Webdesign Toolbox

Webdesign Toolbox is a collection of tools, apps, and resources all in one location for designers and developers. The best part of this resource is that it is human-curated, so every tool is quality checked and makes the list because it has been tested and researched. Search the collection by design, dev, stock, typography, UX, or workflow tools (and more) and use them to help create more efficiently. The collection is constantly growing, too.

CodeStream

CodeStream might be the new-world workflow tool for web designers and developers. It is made for remote teams to review code right inside your IDE without breaking from development flow. You can post and review changes and comments are all independent of the code itself, even though they link to it.

Litur

Litur is a color management app for iOS. Use it to find and collect color swatches, create custom palettes, and even check color combinations against accessibility standards. The app can even generate color codes for you from swatches you find from a photo or image upload or create. The app works on mobile and desktop Mac devices and is a paid app.

Editor X

Editor X, which is still in beta, is a website building tool that combines advanced design and prototyping capabilities with secure web hosting and integrated business solutions. Go from an idea straight to production in a highly intuitive design workspace. The best feature might be exact design precision tools.

Grid Cheatsheet

Grid Cheatsheet is a visual and code-based set of “cheats” based on the W3C CSS Grid Specifications. What’s nice is it makes these guidelines easier to understand and use if reading through them makes you a little uneasy.

Tutorialist

Tutorialist brings together some of the best development tutorials on the web. All of the tutorials are free videos available on YouTube, and this project collects them all in one place.

Pure CSS Halftone Portrait from JPG

Pure CSS Halftone Portrait from JPG is a beautiful pen from Ana Tudor that shows how to change the visual representation of an image. The examples are brilliant and in true halftone fashion. The code snippet works with color, or black and white images as well.

VoiceText for Slack

VoiceText for Slack is another work from home productivity tool. Integrate it with Slack and send messages with text that’s transcribed right in your channels. It’s a free integration and supports 18 languages.

Feature Peek

Feature Peek is a developer tool that helps you get frontend staging environments on demand and gather team feedback earlier in the development process. It’s made for use with GitHub and works with a variety of other tools as well.

Formbutton

Formbutton is a simple and customizable pop-up form. (And we all know websites have plenty of them right now.) It connects to other services you already use, such as Google Sheets and MailChimp, and is simple to set up.

Blocksy Theme

Blocksy is a WordPress theme that’s made for non-coders. It’s a zippy and highly visual theme made for Gutenberg. It works with other builders and allows the user to customize pretty much everything visually. (There’s even a dark mode.) The theme is packed with tools and options and is a free download.

Oh My Startup Illustrations

Oh My Startup Illustrations is a set of vector illustrations in several categories featuring a popular style on many projects. Use the characters and scenes to create a semi-custom story for your startup project.

1mb

1mb is a code editor and host where you can create a static website with a custom domain and SSL included. The editor works in-browser and everything is saved in the cloud.

Linear

Linear is an issue tracking Mac app for teams. It’s designed to help streamline software projects, sprints, and tasks, and can integrate with standard tools such as Github, Figma, and Slack.

Hosting Checker

Hosting Checker solves a common issue – a client wants you to work on their website, but has no idea who hosts it. Hosting Checker shows the user hosting provider and IP address the website uses, along with where its server computers are located and the host’s contact details. It also claims to be 82% faster than other similar tools.

Spike

Spike alerts you to website incidents before customers. Create alerts and get a phone call, text message, email, or Slack notification right away. The tool provides unlimited alerts and integrations to you can stay on top of issues before they become real problems.

Magnus UI

Magnus UI is a framework that helps you building consistent user interfaces in React. It comes with plenty of components ready to use and you can customize the theme.

SpreadSimple

SpreadSimple uses data in Google Sheets to create styled websites with features such as filtering, search, sorting, cart, order collection via forms, and much more. Update the sheet and instantly see changes on the website.

WebP vs. JPEG

Google is starting to suggest using it’s WebP image format to decrease load times, because of the lighter file size. But is WebP better than the traditional JPEG? Developer Johannes Siipola tested the file types at different sizes to answer the question. The answer is a bit complicated, but sometimes it might be better; read the full analysis for more.

Oh Dear

Oh Dear is a website monitoring tool that can help you keep a check on websites. Monitor uptime, SSL certificates, broken links, and more with notifications that come right to you if there’s an issue.

Airconnect

Airconnect is a Zoom video conferencing alternative that you can use for your brand with a custom header, colors, and portal for clients. The tool includes video calling as well as the ability for customers to access their data and automate your onboarding process.

Free Faces

Free Faces is a curated collection of free typefaces that you can browse and use in projects. Search by type style with visual results that include a download link.

All the Roll

All the Roll is a fun novelty font for just the right type of project. It includes 167 characters with swash characters that can be added before or after certain letters.

Backrush

Backrush is a handwriting-style typeface with easy strokes and a pen-like feel. It includes thicker letterforms with nice swashes and a full character set.

Thuner

Thuner is a slab display font with interesting quirks. It’s made for larger than life designs. It includes a full uppercase character set and numerals.

Source


Source de l’article sur Webdesignerdepot

Deep Learning has spurred interest in novel floating point formats. Algorithms often don’t need as much precision as standard IEEE-754 doubles or even single precision floats. Lower precision makes it possible to hold more numbers in memory, reducing the time spent swapping numbers in and out of memory. Since this where a lot of time goes, low precision formats can speed things up quite a bit.

Here I want to look at bfloat16, or BF16 for short, and compare it to 16-bit number formats I’ve written about previously, IEEE and posit.


Source de l’article sur DZONE (AI)