Alternatives to ChatGPT: free, open source…
We talk a lot about ChatGPT, it’s even the name that comes up most often when we talk about textual AI and chatbot. However, other projects have emerged or are in development. Here are some alternatives to ChatGPT, some of which may well overshadow it.
The trend ” technologyof 2023, these are the so-called artificial intelligences generative capable of producing images or text. Among them, the best known is unequivocally ChatGPT, an AI chatbot developed by the OpenAI company. But it is far from being the only one: in recent months, projects have multiplied. Some are fully launched, others under development: here are the alternatives to ChatGPT.
Bing Chat: the same system as ChatGPT, but integrated everywhere
At the beginning of February 2023, Microsoft officially entered the generative AI race by transforming Bing, its search engine. The latter took advantage of the integration of the GPT-4 language model almost immediately. It is this AI model, developed by OpenAI, that is found in the paid version of ChatGPT. The chatbot, in free version, uses GPT-3.5, a previous version of the model, less creative and less powerful.
Since then, Microsoft has integrated Bing Chat everywhere: the tool has naturally invited itself into Edge, in its Android Swift Key keyboard, but not only. Subsequently, it will integrate natively in Windows 11, although for the moment it will be necessary to go through unofficial applications like Easychat AI.
Less efficient than ChatGPT, but connected to the Internet
Although Bing Chat uses the same model as ChatGPT, there are concrete differences between the two chatbots. First of all, Bing Chat is less efficient than its ” competitor“.
Bing Chat is less powerful. This choice was made for economic reasons perhaps, but officially for security: it avoids the excesses of “new Bingeven though the stakes are high.
His answers are generally shorter and less exhaustive. They relate more to the synthesis of search results than to the generation of text. This is especially noticeable in the taskscreativelike writing a letter, or in more pragmatic tasks: calculation, computer programming. A completely normal observation: this is the objective of Bing Chat.
The main advantage of Bing Chat compared to ChatGPT is that it is “connected to internet» : this means that we can deal with recent information. In addition to that, he cites his sources: each information given links to a web page.
Google Bard: the web giant has its AI
Faced with competition from OpenAI and indirectly from Microsoft, Google was pushed to react. Last February, the web giant presented Bard, a conversational robot that aims to compete with ChatGPT. It is based on the LaMDA language model (Language Model for Dialogue Application) and will be integrated directly into the Google search engine, but also on Chromebooks or in the Google application.
At the moment, Bard is available in 180 countries, but not in the European Union. In question: the legislation of the Union precisely, too restrictive in terms of the protection of personal data for Google. Fortunately, there is a method to use Bard in France. This is not the only limitation of the chatbot: for the moment, we can only converse with it in English, for security measures.
Bard isn’t the only tool Google is working on. After having invested nearly 400 million dollars in the start-up Anthropic, the company is developing with the latter Claude, yet another AI chatbot. Its stated competitive advantage is the fact that it is less “dangerousin its results.
ChatGPT vs Google Bard: which is the best chatbot?
If Google Bard was quite basic in its infancy, the chatbot is gradually being completed through regular additions: improving sources and text summaries, displaying images in responses, taking taking into account its location or even the creation of tables that can be used directly in Sheets.
On the length of answers, ChatGPT and Google Bard are on the same level overall. Bard comes across as quite loquacious and thorough in most of his responses. The addition of images in the answers is a real plus, which adds a welcome formatting. It is sometimes worse than ChatGPT, especially on text generation. For now, Bard is kind of in between Bing Chat and ChatGPT: sourced, formatted, but less good at pure text generation. A balance that could well revolutionize Google, as a search engine.
Meta is doing it too
Microsoft and Google aren’t the only Gamam getting into generative text AI. In August 2022, Meta created and released BlenderBot 3.0, an AI chatbot that also had internet access. But by dint of relaying conspiracy theories and criticizing Mark Zuckerberg himself, the tool had been deleted a few days after its launch.
The company resumed its work and developed and released LLaMA in February, its own language model. Unfortunately, the latter saw its source code leaked, so it ended up in the wild. One developer even managed to run LLaMA locally on his computer.
For its part, Meta claims to take into account 65 billion parameters during generation. Its language model is said to outperform GPT-3 in several aspects. Leaks have shown that an AI chatbot could be coming to Instagram, akin to My AI, the assistant “cleverfrom Snapchat, which has experienced significant abuses. However, LLaMA cannot be used by the general public: for the moment, only certain scientists have access to it.
ChatSonic: based on OpenAI’s GPT, but more intuitive than ChatGPT
Another well-known AI chatbot is ChatSonic, which automatically presents itself as an alternative to ChatGPT, aChatGPT like. Templates,quickpre-configured, browser extension, etc. All this makes this chatbot more intuitive than ChatGPT, even in its free version.
In fact, it appears to be less efficient than ChatGPT for text generation, but has other advantages. First of all, the latter can be based on research results, which makes it possible to have up-to-date answers. Also, you can generate images (via Stable Diffusion) or use voice synthesis to dictate your request.
ChatSonic has “personalities», you can select one according to what you want to do: translation, travel guide, poem, astrology,consulting, accounting, etc. Enough to chew up the work of query engineering and go faster. Attention all the same: the free version limits to 10,000 words generated per month (to be divided by two if your requests take into account the updated search results).
Other projects similar to ChatGPT: when search engines dedicate themselves to it
Chinese online search giant Baidu is also rumored to want its ChatGPT-style search engine. More generally, China is also pushing for the development of language models and would have the ambition to rise to the level of American models. We also note Brave Search and DuckDuckGo, which are timidly embarking on generative AI.
We should also mention You, a search engine that is ahead of Google Bard and Bing Chat, with its YouChat function, a chatbot that also integrates Stable Diffusion, an image generation tool, as well as graphics, tables, etc.
The paid version of You also allows you to use the GPT-4 model and generate complete texts. As for the free version that we tested, it works quite well: the answers are often more complete than Bing Chat, but less sourced. You is more creative and allows you to do more things, as mentioned above. In terms of “performancepure, it is worse than ChatGPT.
Open source language models
GPT, LaMDA, LLaMA, etc. are language models of which we do not know the depths, since they are private initiatives. However, public research has taken up the issue with several AI initiatives like ChatGPT.
Bloom: the most French language model
BLOOM is arguably the largest public language model ever developed, released in November 2022, in collaboration with HuggingFace. It is a New York startup founded by three French people. One of the reasons for the creation of Bloom is precisely to give power back to everyone by avoiding that LLMs are only private. It is a model using 176 billion parameters, trained with data from 46 languages.
A model created by BigScience, an association bringing together hundreds of researchers around the world, and trained for three and a half months by Jean Zay, the French public supercomputer. However, it’s hard to find a Bloom-based chatbot that really works. For the moment, this LLM is only in its infancy.
Other open source models
Among the other initiatives developed, we can mention Dolly 2.0 from Databricks. An LLM that wants to be secure, both in terms of data protection and what the chatbot can write. At the moment it is only a model and there is no easily usable chatbot online.
Another recently released model: StableLM, from Stability AI, best known for Stable Diffusion, the image generator. Also in open source, it takes into account up to 7 billion parameters, but Stability AI would like to go up to 65 billion in the near future. In fact, it is much less efficient than ChatGPT: its answers are sorely lacking in logic and are too often wrong. In addition, the chatbot really struggles to manage contexts: if you don’t clean up a conversation, it can stay on the first idea given.
In fact, this is not exactly the philosophy of StableLM: this LLM is more designed to be a basis for chatbots performing specific tasks. Especially since the example above is in French: the model is much more useful in English. You can still try it on the HuggingFace platform since Stability AI has made its model available.
To follow us, we invite you to download our Android and iOS application. You can read our articles, files, and watch our latest YouTube videos.
