As LLMs have grown more powerful, hallucinations have proven stubbornly difficult to avoid. Errors pop up in even the smartest models, and while there are ways to catch those errors, the industry is still figuring out the best way to do it.<\/p>\n
Probably, which just raised $9 million in seed funding from Andreessen Horowitz, is trying to build a more rigorous way to catch those errors.<\/p>\n
As founder Peter Elias (pictured above) puts it, the company\u2019s goal is to prevent hallucinations and simple factual errors from ever reaching the user, and achieve the kind of 99.99% accuracy that\u2019s common in deterministic systems but much more difficult to reach with AI. As it turns out, bringing LLMs to that level of accuracy requires rethinking many of the basic assumptions of AI engineering.<\/p>\n
Probably\u2019s first product is a data science tool, built to produce quick answers from complex datasets. Each result comes with a citation and an audit trail for how it was developed, an increasingly common practice among AI tools.<\/p>\n
But keeping errors from creeping into those summaries required an elaborate harness system that Elias describes as a \u201cdata science mech suit.\u201d The LLM\u2019s first-pass answers are checked against a deterministic validator system, which bounces back any results that don\u2019t match the dataset. Crucially, the LLM has been trained against the validator, and the whole system is optimized for fast and accurate answers, the company said.<\/p>\n
\u201cWhat we learned building this was that the better your harness engineering is, the weaker the model can be,\u201d Elias says. \u201cIf you can refine the context enough, the model does not have to work very hard to do the right thing. Basically, it\u2019s an exercise in reducing ambiguity.\u201d<\/p>\n
That allows Probably\u2019s data science tool to run on significantly smaller AI models. Elias says the current version is running on a model that\u2019s \u201cfour classes weaker than the frontier models,\u201d which means it can be run on local hardware (that is, a desktop computer instead of a data center), which reduces a huge amount of the token costs associated with AI use.\u00a0<\/p>\n
It\u2019s a welcome idea at a time when token costs are rising and many customers are reassessing their AI budgets. And, Elias\u2019 idea doesn\u2019t end with data science, as the same engine can be extended to cover use cases like accounting or medical services \u2014 as Elias puts it, \u201cany precision-sensitive use case.\u201d<\/p>\n
\u201cI think it\u2019s really interesting that the big AI labs have not even attempted to do this,\u201d Elias says. \u201cThey\u2019re incentivized not to, because they make money the more times you have to correct the model.\u201d<\/p>\n<\/div>\n
[analyse_source url=”https:\/\/techcrunch.com\/2026\/06\/16\/probably-raises-9m-to-build-a-more-reliable-kind-of-ai\/”]<\/p>\n","protected":false},"excerpt":{"rendered":"
[analyse_image type=”featured” src=”https:\/\/techcrunch.com\/wp-content\/uploads\/2026\/06\/probably-peter-elias.png?resize=1200,800″] As LLMs have grown more powerful, hallucinations have proven stubbornly difficult to avoid. Errors pop up in even the smartest models, and while there are ways to catch those errors, the industry is still figuring out the best way to do it. Probably, which just raised $9 million in seed funding from […]<\/p>\n","protected":false},"author":1,"featured_media":0,"comment_status":"open","ping_status":"","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[2],"tags":[226,62],"class_list":["post-1994343","post","type-post","status-publish","format-standard","hentry","category-politics","tag-crawlmanager","tag-techcrunch-com"],"_links":{"self":[{"href":"https:\/\/analyse.optim.biz\/index.php?rest_route=\/wp\/v2\/posts\/1994343","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/analyse.optim.biz\/index.php?rest_route=\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/analyse.optim.biz\/index.php?rest_route=\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/analyse.optim.biz\/index.php?rest_route=\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/analyse.optim.biz\/index.php?rest_route=%2Fwp%2Fv2%2Fcomments&post=1994343"}],"version-history":[{"count":0,"href":"https:\/\/analyse.optim.biz\/index.php?rest_route=\/wp\/v2\/posts\/1994343\/revisions"}],"wp:attachment":[{"href":"https:\/\/analyse.optim.biz\/index.php?rest_route=%2Fwp%2Fv2%2Fmedia&parent=1994343"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/analyse.optim.biz\/index.php?rest_route=%2Fwp%2Fv2%2Fcategories&post=1994343"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/analyse.optim.biz\/index.php?rest_route=%2Fwp%2Fv2%2Ftags&post=1994343"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}