Daniel Nicorici

Bioinfo

[Daniel Nicorici @dan2nic]

@JiaJennyLiu No. Most academics are trained to falsify an hypothesis, and see what hyphotesis survive this process. Looking for evidence means to truthify the hypothesis which is not scientific in any shape or foem. Story telling has big role but should talk about science.

[Daniel Nicorici @dan2nic]

@artetxem Suicide by AI?

[Daniel Nicorici @dan2nic]

@woodchipdaddy @lpachter because biologists will want to claim ownership of it, probably it will be called "mathematical biology" (see "computational biology")

[Daniel Nicorici @dan2nic]

@5_utr @LocasaleLab Plus the already published life science articles have like 50% ireproducible results on which AI feeds on as truths

[Daniel Nicorici @dan2nic]

@RuxandraTeslo @nytimes Are you kidding me? At least 50% of results published in life science articles are not reproducible. Why these unreproducible results would be allowed to move towards clinical trials? It is very ezpensive to validate the results feom these articles

[Daniel Nicorici @dan2nic]

@tangming2005 @grok Isn’t this historically and statistically misleading? Classical statistics was not built for p≫n datasets. edgeR & DESeq2 were developed to make small-n RNA-seq inference possible by borrowing information across genes.A naive t-test plus volcano plot is not the equivalent

[Daniel Nicorici @dan2nic]

@panomics @nimivashi15 @doctorveera What are you smoking? Eroom law says otherwise.

[Daniel Nicorici @dan2nic]

@Dr___Tsai @theannasu Exactly

[Daniel Nicorici @dan2nic]

@theannasu No.

[Daniel Nicorici @dan2nic]

@Oliver__Hahn @claudeai @AnthropicAI Open AI all the way

[Daniel Nicorici @dan2nic]

@nc_frey I guess that this is how research will be done with AI from now on.

[Daniel Nicorici @dan2nic]

@PI010101 @pozsgaybalazs But this is not science. It some fraternity or hobby club.

[Daniel Nicorici @dan2nic]

@nguyentony_ @stuartbuck1 I call this truthify the hypothesis.

[Daniel Nicorici @dan2nic]

@nguyentony_ @stuartbuck1 Unfortunately this is the future. Even before AI the reproducibility rate for results published in life science articles was around 50%. Now with AI will be around 1%.

[Daniel Nicorici @dan2nic]

@stuartbuck1 Imagine when people will use AI to fabricate data for research.

[Daniel Nicorici @dan2nic]

@samth Speed cameras are validated independently. Pangram is not.

[Daniel Nicorici @dan2nic]

@tanelitikka Public institutions do not prevent all companies from succeeding. They selectively reduce risk for institutionally recognized (big/old) companies while leaving independent/small investors and challengers to absorb their losses privately. => big ones are favoured

[Daniel Nicorici @dan2nic]

@scserendipity1 @seanmmcbride @Prof_IanD But Panagram is not validated independently. Not sure that is the best or the worse. Use at least an ensemble of three AI detectors to make an opinion.

[Daniel Nicorici @dan2nic]

@s_batzoglou @YourGoldDwarf @KristapsBalodi3 But who will have time to check all the results in arxiv? There is not enough human power for this.

[Daniel Nicorici @dan2nic]

@LaplaceTransf17 @KristapsBalodi3 Probably in the future only artickes where authors which have license for AI for a field of science, AI for math, will be allowed only to publish. One can still use AI freely without a license for math but will not be able to publish.