# Science ChatGPT Plugins

> Source: https://aiwiki.ai/wiki/science_chatgpt_plugins
> Updated: 2026-05-16
> Categories: AI for Science, ChatGPT
> From AI Wiki (https://aiwiki.ai), a free encyclopedia of artificial intelligence. Quote with attribution.

Science [ChatGPT plugins](/wiki/chatgpt_plugins) were a group of third-party tools that extended [ChatGPT](/wiki/chatgpt) with the ability to perform exact computation, fetch peer-reviewed literature, run symbolic mathematics, retrieve curated scientific data, and read structured databases of scholarly papers. The group was active from the platform launch on March 23, 2023, until [OpenAI](/wiki/openai) shut the plugin system down on April 9, 2024. During those twelve and a half months, the science segment was one of the most discussed parts of the plugin store, in large part because the inaugural launch already included Wolfram, the partner most closely associated with quantitative scientific work.[1][2]

This article is a historical reference. The plugins below can no longer be installed or invoked through the original plugin runtime. Most of the underlying products are still available either as standalone web services, as [Custom GPTs](/wiki/custom_gpts) inside the [GPT Store](/wiki/gpt_store), or as actions that other tools call directly.[14]

## scope of the science category

OpenAI never published an official taxonomy for the plugin store, so the boundary between "science" and adjacent groupings such as research, education, and data was always loose. In practice, observers and aggregator sites placed a plugin in the science group when its primary purpose was one of the following:

* exact mathematical, physical, or chemical computation,
* retrieval of peer-reviewed scientific literature with real metadata,
* access to curated scientific data feeds such as astronomical positions, weather, or NASA imagery,
* visualization of scientific concepts through diagrams and charts,
* parsing and question answering against the PDF copy of a research article.

Many of the same tools were also catalogued under [academic research ChatGPT plugins](/wiki/academic_research_chatgpt_plugins) and [space ChatGPT plugins](/wiki/space_chatgpt_plugins). The current article focuses on the science framing, which leans toward computation and the natural sciences rather than humanities scholarship or generic productivity. The boundary mattered in practice because users typically had a hard cap on how many plugins could be enabled in a single conversation, so they had to pick the small handful most relevant to a given task.

For context on how the broader store was organized, see [ChatGPT plugin categories](/wiki/chatgpt_plugin_categories).

## platform timeline

| Date | Event |
| --- | --- |
| March 23, 2023 | OpenAI announces the ChatGPT plugin system; the initial set of twelve third-party launch partners is Expedia, FiscalNote, Instacart, KAYAK, Klarna, Milo, OpenTable, Shopify, Slack, Speak, Wolfram, and Zapier, plus the two first-party plugins for web browsing and code execution. Wolfram is the only science-focused launch plugin.[1][2] |
| March to April 2023 | The system is in private alpha; roughly seventy plugins reach early developers and a waitlist opens for ChatGPT Plus subscribers. Wolfram is highlighted as a flagship example for math and science.[7] |
| May 12 to 18, 2023 | The plugin store and browsing beta open to all ChatGPT Plus subscribers in a staged rollout. The catalog reaches roughly seventy plugins in the first wave and grows past two hundred within weeks, exposing science tools such as ScholarAI, AskYourPDF, OA.mg Science, and Show Me Diagrams to the wider audience.[3][15] |
| Mid 2023 | Diagram, computation, and literature plugins proliferate; aggregator sites begin grouping them informally as a science category. Show Me, Link Reader, WebPilot, AskYourPDF, and ScholarAI emerge as the most frequently recommended tools for researchers.[12][16][17] |
| September 2023 | PubMed Research and similar biomedical plugins are added to the store, expanding the science group into life sciences.[18] |
| November 6, 2023 | OpenAI announces [Custom GPTs](/wiki/custom_gpts) at DevDay, beginning the migration away from plugins. Consensus debuts a Consensus GPT later the same month.[4][19] |
| January 10, 2024 | The [GPT Store](/wiki/gpt_store) opens to ChatGPT Plus, Team, and Enterprise subscribers.[5][20] |
| March 19, 2024 | The plugin store closes; new installations and new conversations with plugins are disabled.[6] |
| April 9, 2024 | All remaining plugin conversations end and the plugin runtime is fully retired.[6] |

## why a science category emerged

The [large language model](/wiki/large_language_model) inside ChatGPT during 2023, which moved from GPT-3.5 to [GPT-4](/wiki/gpt-4) over the year, had two persistent weaknesses for science work. Its training data had a fixed knowledge cutoff, so it could not see recent papers, and it could produce numerical answers that looked correct but were arithmetically wrong, including invented citations. Plugins addressed both problems by routing specific kinds of questions to external services that returned authoritative results. The user typed a question in natural language, the model decided which plugin to call, the plugin returned structured data, and the model then phrased a final answer.

This architecture was the production debut of what later writers would describe as tool use through structured calls, and the science category was the most visible early showcase. The emergence of this style of interaction also pushed practitioners to refine their [prompt engineering](/wiki/prompt_engineering) for hybrid model and tool workflows. From a system design standpoint, each plugin was defined by a small ai-plugin.json manifest that pointed to an OpenAPI specification. ChatGPT read that specification, decided which endpoints were relevant to a user query, generated a JSON request body, and parsed the JSON response. That same pattern of manifest plus OpenAPI later became the foundation for GPT Actions, which kept the protocol largely intact while changing the surface that users interacted with.[14]

### the hallucination problem in science

The scholarly literature was unusually vocal about the cost of getting plugins right for science use. A widely cited 2023 article in the journal Scientific Reports tested ChatGPT on bibliographic generation and found that of 115 references the model produced, only 7 percent were both authentic and accurate, while 47 percent were entirely fabricated and 46 percent were authentic but inaccurate.[21] Subsequent work in economics, psychiatry, and clinical medicine reported similar fabrication rates, ranging from about 20 percent for GPT-4 to over 50 percent for GPT-3.5.[22] Plugins that grounded an answer in real metadata, returning a digital object identifier, an actual title, and a verifiable URL, were the most direct mitigation available before retrieval augmented generation became standard.

## Wolfram

Wolfram was the centerpiece of the science category and the only science-focused plugin in the original launch lineup of March 23, 2023. Stephen Wolfram announced the integration on the same day, describing it as giving ChatGPT "computational superpowers" by routing queries to Wolfram|Alpha and the Wolfram Language.[7]

The plugin covered an unusually wide subject area for a single tool. It handled:

* mathematics, including symbolic algebra, calculus, equation solving, integrals, and plotting,
* physics constants, unit conversions, and formula evaluation,
* chemistry, including element properties, molecular formulas, and reaction stoichiometry,
* astronomy, including planetary positions, distances, and the location of the International Space Station,
* geography and earth science, including distance computation, projections, and choropleth maps,
* biology and anatomy reference data,
* statistics, data visualization, and code generation in Wolfram Language.

When a user asked a quantitative question, ChatGPT translated the prompt into a Wolfram|Alpha query, sent it to the plugin, and read the structured response. The model then composed a sentence-level answer, often embedding plots or maps that Wolfram|Alpha returned. Wolfram described the partnership as a meeting between symbolic and statistical traditions of artificial computing, with [Wolfram Alpha](/wiki/wolfram_alpha) supplying exactness and ChatGPT supplying conversational framing.[7]

A peer-reviewed evaluation by Ernest Davis and Scott Aaronson posted on arXiv in 2023 tested GPT-4 on 105 original problems in math and science at high school and college level, with and without the Wolfram and Code Interpreter plugins. The authors reported that the plug-ins materially improved performance on quantitative tasks compared with the model alone, but that GPT-4 frequently failed to take full advantage of Wolfram, performing computations itself rather than handing them off and stumbling on the interface between natural language and the formal Wolfram Language. The systems were strongest on single-formula problems, often weak on tasks requiring spatial visualization, and often weak on multi-step problems that combined several different kinds of calculation.[8]

In April 2023, Wolfram followed the plugin with a separate Wolfram ChatGPT Plugin Kit that let developers wrap Wolfram Language code as their own plugins. The kit allowed anyone with a Wolfram Cloud account to publish a custom plugin without writing the OpenAPI scaffolding by hand.[9]

## ScholarAI

ScholarAI was the most widely cited literature search plugin in the science group. It let ChatGPT query a corpus of more than 200 million scientific texts indexed across PubMed, arXiv, and Springer-Nature, returning titles, authors, publication dates, abstracts, and direct links to PDFs where available.[10] The plugin could also fetch the full text of a paper from a supplied URL and let the user ask follow-up questions about specific sections, figures, and tables.[10]

ScholarAI was promoted as a way to ground responses in real, traceable references rather than the fabricated citations that the base model sometimes produced. Each answer surfaced through the plugin included the underlying paper's metadata, which made cross-checking straightforward. The service worked best for biomedical and life sciences questions because of its strong Springer-Nature and PubMed coverage, but it also returned relevant preprints from arXiv for physics, computer science, and mathematics.

After the plugin sunset, ScholarAI migrated to a Custom GPT of the same name in the [GPT Store](/wiki/gpt_store). The product continues to maintain the underlying search index and PDF question answering as a standalone web app at scholarai.io, and the company also exposes its search through actions and an MCP server that other AI clients can call.

## Consensus

Consensus launched its ChatGPT plugin in 2023 against the same Semantic Scholar derived corpus that powers its standalone search engine, advertised as more than 200 million peer-reviewed papers.[19] The plugin focused on answering yes-or-no science questions and short factual queries by surfacing the actual study text and a synthesized verdict. Unlike a general literature search tool, Consensus emphasised summarisation: it presented the percentage of papers that found a positive, negative, or mixed result on the question, with citations the user could click through.

Consensus rebuilt the same product as Consensus GPT in November 2023, shortly after OpenAI announced Custom GPTs at DevDay, and that GPT remains one of the most installed science research tools in the GPT Store.[19] The company has since extended the same index to a model context protocol server so that other AI clients can query the corpus directly.

## OA.mg Science

The plugin published by OA.mg, listed in the store simply as Science, used semantic search across a database of roughly 250 million scientific papers and research articles.[11] It extracted keywords from the user's natural-language question, then ranked results by a combination of textual similarity and citation count, so highly cited papers tended to appear first when relevance was otherwise tied.[11] The metadata returned included title, authors, publication year, key concepts, and abstract.

OA.mg promoted the plugin as a way to bring open-access scientific literature into ChatGPT and noted that users could find and install it by searching for "Science" in the plugin store.

## PubMed Research

PubMed Research was added to the plugin store on September 22, 2023 and quickly became the recommended biomedical literature plugin alongside ScholarAI.[18] It connected ChatGPT directly to the National Library of Medicine MEDLINE corpus and let users filter by date range, number of results, and standard PubMed search syntax. Returned items included the PubMed identifier, the journal, the publication date, and the abstract, so the model could compose grounded summaries of clinical or life sciences questions without inventing references.

A family of related biomedical plugins, including SightBot and several Paperpile derived tools, exposed the same MEDLINE data through slightly different interfaces, often combining PubMed search with Semantic Scholar, Crossref, and Springer-Nature endpoints.

## AskYourPDF and ChatWithPDF

AskYourPDF and ChatWithPDF turned uploaded or linked PDF files into a searchable corpus that ChatGPT could read. Although both plugins were widely used outside science, they were staples of the scientific workflow because so much primary literature is distributed as PDF. Users pointed the plugin at a research paper, lecture notes, or scanned slides and asked questions whose answers required reading specific sections, equations, or figures of the document.[12][17]

AskYourPDF returned answers with the page number of the cited passage, which made it easy to verify the model's claim against the original. ChatWithPDF accepted both file uploads and direct URLs and added basic translation and rewriting features on top of the question answering loop. The ChatGPT Plus subscription was required to use both, since plugin access was tied to that paid tier throughout the platform's life.

For longer documents, users typically chained the workflow: an article search through ScholarAI, Consensus, or OA.mg returned a candidate paper, the plugin output included a PDF URL, then AskYourPDF or ChatWithPDF parsed that PDF in place. The combination was the practical replacement for the older pattern of pasting full article text into the chat, which often hit context length limits.

## Link Reader and WebPilot

Link Reader and WebPilot were general purpose web readers that overlapped with the science workflow whenever a paper lived on a website that ScholarAI or PubMed did not index. Link Reader read web pages, PDFs, PPTs, images, Word documents, and similar formats and summarised them inside the chat. For research papers it could extract key points and figures from PDFs that the user supplied as URLs, including preprints on lab websites and conference proceedings.[16]

WebPilot offered a similar surface focused on web URLs and dynamic web data. Reviewers consistently grouped both plugins among the most useful research tools because they bridged the gap between general web browsing, which OpenAI's first-party browsing plugin handled, and the structured paper search that ScholarAI and Consensus provided.

## diagram and visualization plugins

Visualization was a common need for science users. Show Me Diagrams, often listed simply as Show Me, let ChatGPT render flowcharts, sequence diagrams, mind maps, and similar figures inline with conversation. The plugin supported Mermaid as the primary syntax and could also output other diagram languages such as GraphViz and PlantUML.[13] In a science context it was used to draw process diagrams for the water cycle, the digestive system, cell division, vaccine development pipelines, and similar concepts that benefit from a quick schematic rather than a long textual description.

Mermaid Chart published its own official ChatGPT plugin in 2023 that combined the same Mermaid syntax with a hosted editor, so users could iterate on a generated diagram outside the chat.[23] Wolfram itself produced certain forms of scientific visualization, including 2D and 3D plots, geographic maps, and rendered chemical structures, returned as images alongside its computational answers.

## space and astronomy plugins

Space tools formed a small but visible subgroup. Space Photo Explorer, distributed through the same plugin runtime, exposed NASA imagery to ChatGPT, including the Astronomy Picture of the Day, the NASA Image Library, and photographs from Mars rovers. Astrodaily offered a similar surface for daily NASA images and tutorials. These plugins gave the model the ability to retrieve specific images on request rather than describe them from training memory.

Wolfram covered the analytical side of space data, returning planetary positions, distances, and other celestial computations. For a fuller treatment of this subgroup, see [space ChatGPT plugins](/wiki/space_chatgpt_plugins).

## ArXiv and preprint plugins

Several plugins targeted arXiv directly. Research By Vector built an embeddings index over arXiv abstracts and let ChatGPT pull semantically similar papers given a candidate title and abstract, which suited the early stages of a literature review where the user does not yet know the right keywords. ArXiv Sanity derived plugins and Paper Interpreter style tools sat in the same niche, with Paper Interpreter focused on rewriting a paper in simpler language so a non-specialist reader could follow it.[24] The combined effect was that ChatGPT could act as a casual reading partner for preprints, summarising abstracts and explaining methods sections without forcing the user to navigate the arXiv site itself.

## first-party tools that complemented the science category

Two plugins built by OpenAI sat alongside the third-party science group and were heavily used in scientific workflows. The web browsing plugin gave the model GET-only access to the open web, which let it pull arbitrary HTML pages, including some that ScholarAI or PubMed did not cover. The Code Interpreter plugin, later renamed Advanced Data Analysis, provided a sandboxed Python environment in which the model could run pandas, numpy, scipy, matplotlib, and similar libraries, read uploaded CSV or Excel files, and return generated images. Davis and Aaronson treated Code Interpreter as the closest natural competitor to Wolfram in their evaluation; the two tools succeeded on overlapping but non-identical problem sets, with Code Interpreter stronger at numerical simulation and Wolfram stronger at symbolic manipulation.[8]

## summary table of notable science plugins

| Plugin | Developer | Function | Status after April 9, 2024 |
| --- | --- | --- | --- |
| Wolfram | Wolfram Research | Symbolic and exact computation across math, physics, chemistry, astronomy, geography, biology | Wolfram GPT in GPT Store; Wolfram|Alpha web service |
| ScholarAI | ScholarAI | Search 200M+ peer-reviewed papers; PDF question answering | ScholarAI GPT; scholarai.io web app; MCP server |
| Consensus | Consensus | Semantic Scholar derived search; yes-or-no answers with citation breakdown | Consensus GPT; consensus.app web app; MCP server |
| OA.mg Science | OA.mg | Semantic search of ~250M open-access papers; citation weighted ranking | oa.mg web app |
| PubMed Research | Independent | Direct MEDLINE search with PubMed syntax | Numerous PubMed GPTs |
| AskYourPDF | AskYourPDF | PDF question answering with page-level citations | AskYourPDF Research Assistant GPT |
| ChatWithPDF | Independent | PDF chat from upload or URL, with translation | Replaced by ChatGPT native file upload |
| Link Reader | gochitchat.ai | Read web pages, PDFs, images, Word, PPT | Discontinued; superseded by ChatGPT browsing |
| WebPilot | WebPilot | Browse and analyse web URLs | WebPilot GPT in GPT Store |
| Show Me Diagrams | bra1nDump | Generate Mermaid, GraphViz, PlantUML diagrams inline | Show Me GPT and Mermaid Chart GPT |
| Mermaid Chart | Mermaid Chart | Diagrams with hosted editor | Mermaid Chart GPT |
| Space Photo Explorer | Independent | NASA imagery, Astronomy Picture of the Day, Mars rover photos | Functionality folded into general browsing GPTs |
| Research By Vector | Independent | arXiv similarity search via vector embeddings | Replaced by various arXiv GPTs |
| Paper Interpreter | Independent | Simplify and explain academic papers | Paper Interpreter GPT |

## relationship to other categories

The science group overlapped heavily with two other informal categories. The literature-focused plugins were also catalogued under [academic research ChatGPT plugins](/wiki/academic_research_chatgpt_plugins), since the same tools served scholars across disciplines. The astronomy and NASA tools sat under [space ChatGPT plugins](/wiki/space_chatgpt_plugins). A few science workflows pulled in plugins from outside the group entirely, such as code-execution sandboxes for numerical work, browser plugins for late-breaking preprints, and citation-formatting tools.

For the full taxonomy, see [ChatGPT plugin categories](/wiki/chatgpt_plugin_categories).

## adoption and uptake

At their peak in early 2024 the plugin store hosted roughly one thousand plugins.[25] Aggregator sites tracking the catalogue counted around 800 to 1,000 entries at the point the system was archived. Adoption was uneven. Public commentary from OpenAI, Zapier, and other partners consistently framed plugins as a useful but ultimately niche feature: most ChatGPT Plus subscribers never installed a plugin, the average installer used roughly two per month, and a small number of power users accounted for a large share of total invocations. The science group followed the same pattern. Wolfram and ScholarAI dominated installation counts within the category, while the long tail of more specialised tools attracted dedicated but small audiences.

The practical limits were threefold. Plugins were only available to ChatGPT Plus subscribers, which cut out the free user base entirely. Each conversation supported only a few enabled plugins at once, so a user had to pre-commit to a small toolkit before asking a question. And the model was responsible for deciding which plugin to call, which it sometimes did poorly, occasionally ignoring an enabled plugin or invoking the wrong one. The combined friction explained why OpenAI ultimately replaced the system rather than iterate on it.

## scientific use cases and workflows

Researchers who did adopt the science plugins typically used them in a small number of recurring patterns:

* literature triage, where ScholarAI, Consensus, OA.mg, or PubMed turned a vague question into a short list of candidate papers with real metadata,
* full-text reading, where AskYourPDF, ChatWithPDF, or Link Reader parsed the PDF of a candidate paper and let the user ask focused follow-up questions,
* synthesis, where ChatGPT composed a draft summary or comparison across several papers based on the structured returns from the search plugins,
* calculation and verification, where Wolfram or Code Interpreter handled the numerical or symbolic work that the underlying model did not do reliably,
* visualisation, where Show Me or Mermaid Chart produced a diagram from a textual description of a process,
* data lookup, where Wolfram answered questions about physical constants, unit conversions, atomic properties, or planetary positions.

Librarians and methodologists writing in 2023 and 2024 generally recommended that plugin output be treated as a leading rather than authoritative result. ScholarAI and Consensus answers were largely trustworthy when the returned digital object identifier resolved to a real paper, but reviewers regularly flagged cases where the plugin returned a real paper that did not actually support the synthesised claim. The pattern resembled earlier debates about Wikipedia in academic work and led to the same recommended practice: use the plugin to find sources, then read the sources directly before citing them.

## limitations and reliability concerns

Alongside the citation hallucination problem already described, the science plugins inherited several other limitations that motivated their eventual retirement. The most important were:

* coverage gaps, where a paper sat outside the indexed corpus and the model fell back to its training data, sometimes without making the distinction clear to the user,
* paywalls, where the plugin returned only a title and abstract and the model attempted to reason about the full paper anyway,
* inconsistent invocation, where the model decided not to call a relevant plugin or called the wrong one,
* opaque reasoning, where the model summarised a returned abstract incorrectly or conflated several returns into a single statement,
* lack of reproducibility, since plugin output depended on live external services and could change between sessions for the same prompt,
* limited file handling, since most plugins worked from URLs rather than direct uploads until later in 2023,
* and the broader concern that AI generated literature reviews could degrade peer review if reviewers leaned on the plugins to draft reports.[22][26]

Journals including Nature and JAMA published guidance during 2023 and 2024 asking authors and reviewers to disclose any use of AI tools in the writing or review process, which implicitly covered the science plugins.[26]

## decline and migration

Uptake of plugins in general was lower than OpenAI hoped. OpenAI announced [Custom GPTs](/wiki/custom_gpts) at DevDay on November 6, 2023 as a more discoverable replacement, opened the [GPT Store](/wiki/gpt_store) on January 10, 2024, then closed the plugin store to new installations on March 19, 2024 and shut the runtime down completely on April 9, 2024.[4][5][6] OpenAI also published a migration guide explaining that existing plugin manifests could be reused as the OpenAPI action specification for a new Custom GPT, which kept developer effort low.[14]

Many science plugin teams rebuilt their products as Custom GPTs. ScholarAI, Wolfram, Consensus, AskYourPDF, WebPilot, Mermaid Chart, and Show Me all maintained equivalent GPTs after the migration. The functionality therefore remained largely available, but the unified plugin store as a single place to discover and combine science tools no longer existed. OpenAI later layered a separate apps in ChatGPT framework on top of the GPT Store, with several of the former plugin developers, including Consensus and ScholarAI, among the first integrations.

### GPT Store successors

| Original plugin | GPT Store successor | Notes |
| --- | --- | --- |
| Wolfram | Wolfram | Same underlying Wolfram|Alpha and Wolfram Language endpoints |
| ScholarAI | ScholarAI | Same 200M+ paper index; PDF chat; also available as web app and MCP server |
| Consensus | Consensus | Same Semantic Scholar derived corpus; later added as a ChatGPT app |
| AskYourPDF | AskYourPDF Research Assistant | Page-level citations; PDF upload and URL parsing |
| Show Me Diagrams | Show Me | Same Mermaid, GraphViz, PlantUML output |
| Mermaid Chart | Mermaid Chart | Hosted editor for generated diagrams |
| WebPilot | WebPilot | Web URL analysis and summarisation |
| Link Reader | discontinued | Functionality largely covered by ChatGPT browsing |
| PubMed Research | several PubMed GPTs | MEDLINE search and abstract return |
| OA.mg Science | oa.mg web app | Standalone semantic search; no first-party GPT |
| Space Photo Explorer | folded into browsing | NASA APIs callable from general purpose GPTs |
| Paper Interpreter | Paper Interpreter | Same plain-language explanation of papers |

## broader context within ChatGPT history

The science plugin story is best read as the first iteration of tool augmented [ChatGPT](/wiki/chatgpt). Before plugins, every ChatGPT answer relied on the parameters of the [large language model](/wiki/large_language_model) itself, which was poorly suited to up-to-date factual lookup and exact computation. The science group made the case publicly that tool use through structured calls could meaningfully fix those weaknesses. The lessons fed directly into later systems: GPT Actions inside Custom GPTs reused the same OpenAPI scaffolding, the Assistants API exposed similar tool calling primitives to developers, and the eventual model context protocol generalised the manifest into a portable interface that any client could speak. Many of the original science plugin developers, including Wolfram, ScholarAI, and Consensus, were among the earliest publishers in each successor channel.

Viewed from late in the timeline, the science plugins look like an early proof of the wider pattern that AI clients would converge on through 2024 and 2025: a frontier model that reasons in natural language paired with a small set of specialised tools that handle the parts of a task the model is structurally bad at. That pattern is now standard across most major chat assistants, and the architectural debt of the original plugin runtime is largely absent, but the catalogue of science tools that emerged in 2023 mapped out the shape of the problem more clearly than any other category in the early store.

## see also

* [ChatGPT plugins](/wiki/chatgpt_plugins)
* [ChatGPT plugin categories](/wiki/chatgpt_plugin_categories)
* [Academic research ChatGPT plugins](/wiki/academic_research_chatgpt_plugins)
* [Space ChatGPT plugins](/wiki/space_chatgpt_plugins)
* [Custom GPTs](/wiki/custom_gpts)
* [GPT Store](/wiki/gpt_store)
* [ChatGPT](/wiki/chatgpt)
* [OpenAI](/wiki/openai)
* [GPT-4](/wiki/gpt-4)
* [Wolfram Alpha](/wiki/wolfram_alpha)
* [Large language model](/wiki/large_language_model)
* [Prompt engineering](/wiki/prompt_engineering)

## references

1. OpenAI, "ChatGPT plugins," March 23, 2023. https://openai.com/index/chatgpt-plugins/
2. TechCrunch, "OpenAI connects ChatGPT to the internet," March 23, 2023. https://techcrunch.com/2023/03/23/openai-connects-chatgpt-to-the-internet/
3. VentureBeat, "OpenAI is rolling out ChatGPT Plugins to Plus subscribers," May 2023. https://venturebeat.com/ai/openai-is-rolling-out-chatgpt-plugins-to-plus-subscribers-heres-why-thats-a-big-deal/
4. OpenAI, "Introducing GPTs," November 6, 2023. https://openai.com/index/introducing-gpts/
5. OpenAI, "Introducing the GPT Store," January 10, 2024. https://openai.com/index/introducing-the-gpt-store/
6. OpenAI Developer Community, "Plugin Store and New Chats With Plugins Closed March 19 2024," 2024. https://community.openai.com/t/plugin-store-and-new-chats-with-plugins-closed-march-19-2024/689877
7. Stephen Wolfram, "ChatGPT Gets Its Wolfram Superpowers!" March 23, 2023. https://writings.stephenwolfram.com/2023/03/chatgpt-gets-its-wolfram-superpowers/
8. Ernest Davis and Scott Aaronson, "Testing GPT-4 with Wolfram Alpha and Code Interpreter plug-ins on math and science problems," arXiv:2308.05713, 2023. https://arxiv.org/abs/2308.05713
9. Stephen Wolfram, "Instant Plugins for ChatGPT: Introducing the Wolfram ChatGPT Plugin Kit," April 27, 2023. https://writings.stephenwolfram.com/2023/04/instant-plugins-for-chatgpt-introducing-the-wolfram-chatgpt-plugin-kit/
10. ScholarAI, product page. https://scholarai.io/
11. OA.mg, "ChatGPT Plugin: Bringing Science to OpenAI's ChatGPT." https://oa.mg/chatgpt
12. AskYourPDF, "The Best PDF Plugin for ChatGPT." https://askyourpdf.com/blog/the-best-pdf-plugin-for-chatgpt
13. Show Me Diagrams plugin source repository. https://github.com/bra1nDump/show-me-chatgpt-plugin
14. drio, "ChatGPT Plugins vs Custom GPTs vs MCP Apps: What Changed and Why." https://www.getdrio.com/blog/chatgpt-plugins-vs-custom-gpts
15. Search Engine Journal, "200+ ChatGPT Plugins Available To Plus Users," May 2023. https://www.searchenginejournal.com/chatgpt-plugins-web-browsing-beta/486785/
16. NerdsChalk, "How to Use Link Reader ChatGPT Plugin to Extract Content From a Link," 2023. https://nerdschalk.com/use-link-reader-chatgpt-plugin/
17. oa.mg, "The 6 Most Useful ChatGPT Plugins for Researchers," 2023. https://oa.mg/blog/the-6-most-useful-chatgpt-plugins-for-researchers/
18. plugin.surf, "PubMed Research plugin information." https://plugin.surf/plugin/pubmed-research
19. Consensus, "Introducing Consensus GPT, Your AI Research Assistant." https://consensus.app/home/blog/introducing-researchgpt-by-consensus/
20. TechCrunch, "OpenAI launches a store for custom AI-powered chatbots," January 10, 2024. https://techcrunch.com/2024/01/10/openai-launches-a-store-for-custom-ai-powered-chatbots/
21. "Fabrication and errors in the bibliographic citations generated by ChatGPT," Scientific Reports, 2023. https://www.nature.com/articles/s41598-023-41032-5
22. Joy Buchanan, Stephen Hill, Olga Shapoval, "ChatGPT Hallucinates Non-existent Citations: Evidence from Economics," 2024. https://journals.sagepub.com/doi/10.1177/05694345231218454
23. Mermaid Chart, "Mermaid Chart's ChatGPT Plugin Combines Generative AI and Smart Diagramming." https://docs.mermaidchart.com/blog/posts/mermaid-chart-chatgpt-plugin-combines-generative-ai-and-smart-diagramming
24. There's An AI For That, "Paper Interpreter." https://theresanaiforthat.com/gpt/paper-interpreter/
25. AIONX, "ChatGPT Plus Plugins Review: What Happened." https://aionx.co/chatgpt-reviews/chatgpt-plus-plugins-review/
26. Nature, "Is ChatGPT corrupting peer review? Telltale words hint at AI use," 2024. https://www.nature.com/articles/d41586-024-01051-2

*See also: [ChatGPT Plugins](/wiki/chatgpt_plugins), [ChatGPT Plugin Categories](/wiki/chatgpt_plugin_categories) and [Science](/wiki/science)*