I've mentioned pit-manager.mit.edu as a source of NetNews FAQ files, and also as a way to find net address of people who've posted to Usenet forums. (Post a "send usenet-addresses/" message to mail-server there.) It still works, but the host name is changing to rtfm.mit.edu. [Mark_Kantrowitz @glinda.oz.cs.cmu.edu, 6/9/93.]

Thanks mainly to Bill Park's activism, Apple has decided to continue publishing the New Inside Macintosh on the Bookmark CD that comes with the quarterly "develop" magazine. The original Inside Macintosh might also be included, if it fits. [Caroline Rose ([email protected]), 6/14/93.] Inside Macintosh is a multivolume reference library describing Mac architecture, system calls, error numbers, and other "stuff" you need to know to develop Mac software. The new series is 15 volumes at about $30 each, so the magazine and CD at $30/year is a bargain. Sign up with [email protected], (800) 877-5548. Well done, [email protected].

Interested in text markup? Jim Warren is working on a way to represent legislative text -- italics, boldface, underscore, strikeout fonts, line and page numbers, indentation and justification, additions, deletions, etc. -- for display on ASCII terminals. "It took me some time to realize the ramifications of italics versus regular font where they were used for current code in some sections, entirely new proposed-code in other sections, and code amendments in other sections." Warren wants a chronological record of modifications to pending bills, further complicating the displays. And he wants it human-readable on 40-character lines, if necessary. Contact [email protected] for his draft proposal. [6/14/93.]

My report of an AAAI article archive was premature. The first papers, from recent symposia only, should be available now from [email protected]. [Daphne Black ([email protected]), 4/14/93. Maury Johnson.]

I've seen interest in Gerald Tesauro's program for learning backgammon solely by playing against itself. The algorithm is based on Richard Sutton's time-difference reinforcement. (An evaluation function applied at each step is used to reward or punish the decision made at the previous step. This has several advantages over waiting for the final win or loss to give feedback to all decision functions.) The backgammon program reached strong intermediate-level play on its own, and strong master play when given an additional set of hand-crafted features. Tesauro's paper on this, "TD-Gammon, A Self-Teaching Backgammon Program, Achieves Master-Level Play," can be FTP'd as tesauro.tdgammon.ps.Z from pub/neuroprose on archive.cis.ohio-state.edu. The paper will appear in Neural Computation. [[email protected], connectionists, 6/1/93.]

Last week's Communique discussed the Prospector expert system. Tod Levitt says that flaws in Prospector's inference method are documented in Yadrick, Perrin, and Vaughan, "Evaluation of Uncertain Inference Models I: Prospector," in Uncertainty in Artificial Intelligence, Vol. 2, Lemmer and Kanal Eds., North-Holland, 1988. The original paper appeared in 1986 in the second Uncertainty in AI workshop. [[email protected], 6/9/93.]